Harry Potter By Balenciaga | Step by Step Tutorial

Harry Potter By Balenciaga | Step by Step Tutorial — PromptJungle

PromptJungle

AI-assisted summary of Harry Potter By Balenciaga | Step by Step Tutorial — original source. Check the original for full context.

Source published:

Key ideas

  • The Balenciaga-style AI meme production workflow relies on a multi-tool pipeline integrating ChatGPT for prompting, Midjourney for visuals, ElevenLabs for voice cloning, and D-ID for animation. (0:00)
  • Midjourney image quality can be maintained by using specific parameters like --ar 3:2 and --v 5, while aesthetic consistency is managed by injecting thematic keywords like 'grotesque' into the prompt. (1:51)
  • D-ID animation tasks are split into two categories: lip-synced audio files for dialogue and 3-second silent loops generated via manual pause-insertion for runway b-roll. (4:02)
  • Final polish in CapCut requires adding 'Camera Shutter' sound effects and 'Flash' transitions on the beat to mimic a runway photography experience, paired with slow-zoom keyframes to animate the static AI portraits. (5:16)

Detailed notes

## Background This tutorial was published on YouTube by the channel PromptJungle on April 3, 2023. According to the page description, it is a free, step-by-step guide explaining how to recreate the viral Balenciaga-style AI meme using a pipeline of ChatGPT, Midjourney, ElevenLabs, D-ID, and CapCut.

## TL;DR The video outlines an end-to-end workflow for creating a 'Harry Potter by Balenciaga' fashion spoof video without paid courses or complex setups. Viewers learn to engineer formatted Midjourney prompts using ChatGPT, upscale the generated portraits via Upscale.media, and clone actor voices using YouTube audio samples in ElevenLabs. The tutorial then demonstrates how to generate both speaking animations and silent runway loops in D-ID before assembling the final cut in CapCut with dynamic zooms, camera flash transitions, and shutter sound effects.

## Atmosphere & style The presentation is crisp, generous, and utilitarian, driven by the creator's philosophy of democratizing viral creative techniques that others monetize behind paywalls. The pacing feels swift and frictionless, systematically unlocking each AI tool with practical workarounds like pause-scripting silent video loops and manual keyframe zooms. Watching it feels like a rapid creative sprint, turning static internet images into stylish, rhythmic, high-fashion satire with minimal overhead.

## Bulletin - [00:00](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=0s) The Balenciaga-style AI meme production workflow relies on a multi-tool pipeline integrating ChatGPT for prompting, Midjourney for visuals, ElevenLabs for voice cloning, and D-ID for animation. - [01:51](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=111s) Midjourney image quality can be maintained by using specific parameters like --ar 3:2 and --v 5, while aesthetic consistency is managed by injecting thematic keywords like 'grotesque' into the prompt. - [04:02](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=242s) D-ID animation tasks are split into two categories: lip-synced audio files for dialogue and 3-second silent loops generated via manual pause-insertion for runway b-roll. - [05:16](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=316s) Final polish in CapCut requires adding 'Camera Shutter' sound effects and 'Flash' transitions on the beat to mimic a runway photography experience, paired with slow-zoom keyframes to animate the static AI portraits.

## Key takeaways - [00:00](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=0s) The workflow is designed to be accessible to beginners by leveraging free-tier versions of standard AI tools. - [00:45](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=45s) ChatGPT is used to generate a consistent 1990s fashion show theme by assigning specific high-fashion attributes to a list of characters. - [01:25](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=85s) Prompt structure is vital for Midjourney; using a consistent template ensures that style parameters like --ar 3:2 and --v 5 remain uniform across the character set. - [01:51](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=111s) If character resemblance is weak, users should simplify the prompt by removing specific gender identifiers or adjusting the descriptive clothing adjectives. - [02:58](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=178s) Upscale.media is recommended to perform bulk 2x upscaling on character stills to ensure high-resolution video output. - [03:24](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=204s) ElevenLabs Instant Voice Cloning requires a clean, high-fidelity audio sample of the target voice to create an accurate speech model. - [03:38](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=218s) MP3 cutter tools are essential for isolating the cleanest segments of audio from source material before cloning. - [04:02](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=242s) D-ID generates talking animations by syncing a portrait with an audio file, using a text-to-video processing engine. - [04:25](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=265s) Silent ambient animation in D-ID can be achieved by using the clock icon to inject 3 seconds of pause, providing reaction shots for the edit. - [04:53](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=293s) Users are advised to secure music licensing through platforms like Epidemic Sound to avoid copyright issues during monetization. - [05:16](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=316s) CapCut is favored for this project due to its mobile-first interface, which allows for precise timing of transitions and audio sync. - [05:53](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=353s) Keyframes are the primary tool for creating a sense of movement within static AI-generated images, adding a 'push-in' effect to improve visual interest.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=0s) Overview and Prompt Generation with ChatGPT

The tutorial opens by breaking down how to create the viral Balenciaga-style AI animations that have taken over social media platforms. While many online creators charge fees for their prompt engineering recipes, the creator commits to providing the entire workflow for free, asking only for a like to help the video reach more viewers. Beginners need not worry, as short introductory tutorials for every software utilized are provided in the video's description. The showcase project chosen to guide viewers step-by-step through the pipeline is a recreation of the iconic "Harry Potter by Balenciaga" meme.

The process begins with generating character descriptions and structured prompts using ChatGPT. Opening a fresh chat session, the first step is to ask the AI for a clean roster of names: "Give me the 10 most popular Harry Potter characters, just the names." Once the roster is generated, the persona instructions are applied. The creator feeds in a tailored prompt directing ChatGPT: “Now, act as a Balenciaga fashion designer with 20 years of experience. I want you to dress each of these 10 characters in a unique Balenciaga piece for a runway show. Just write the name of the character and the piece. Remember, it's a 1990 fashion show.”

After ChatGPT pairs every character with its high-fashion runway ensemble, it must be trained on the exact output format required by Midjourney. The user provides a master template prompt: “Ok, now this is a prompt that I use to generate images from another AI image generation tool called Midjourney: '1990 screengrab of female model Hermione Granger wearing a grotesque black Balenciaga dress, fashion movie scene, Balenciaga commercial --ar 3:2 --v 5'. Just say 'Y' if you understand.” Once ChatGPT confirms, a follow-up prompt instructs it to rewrite this exact format for all ten characters, maintaining the parameters (`--ar 3:2 --v 5`) while adapting only the gender, character name, and runway attire.

```mermaid flowchart TD accTitle: AI Video Production Pipeline accDescr: Step-by-step flow from ChatGPT prompt creation down to the final edit in CapCut.

subgraph PreProduction ["Pre-Production"] A["ChatGPT: Generate Characters & 1990s Runway Outfits"] B["ChatGPT: Format Midjourney Prompts (--ar 3:2 --v 5)"] A --> B end

subgraph AssetGeneration ["Asset Generation"] C["Midjourney: Image Generation & Selection"] D["Upscale.media: 2x Batch Upscaling"] E["YouTube to MP3 + MP3 Cutter: Voice Extraction"] F["ElevenLabs: Instant Voice Cloning & Speech Synthesis"] C --> D E --> F end

subgraph AnimationAudio ["Animation & Assembly"] G["D-ID: Talking Animation (Audio Upload)"] H["D-ID: Silent Poses (0.5s Pause Script Blocks)"] I["CapCut: Assembly, Flashes, Audio FX & Dynamic Zooms"] D --> G D --> H F --> G G --> I H --> I end ```

[01:51](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=111s) Midjourney Image Generation and Upscaling

With the formatted prompt list ready, the workflow moves to Discord to generate the base portraits via Midjourney. Heading into one of the `rookie` channels, the creator executes `/imagine` and pastes the prompts one after the other. When rendering stylized characters, likeness and aesthetic consistency can occasionally drift. If Midjourney generates faces that do not sufficiently match the original actor or character traits, the creator explains: “If you start getting images that don't resemble the character's facial features, you can always delete the part where it says male or female model.” Similarly, if the clothing veers too far from the desired mood, users can return to ChatGPT and guide the aesthetic by injecting specific thematic keywords like "goth", "grotesque", or a constrained palette such as "only black and white".

Curating the generated results requires creative foresight: “Always envision the video while you are choosing your images for the best outcome and a higher quality content.” After picking the best generations and upscaling them within Discord, some outputs may still lack crisp detail for video production. To solve resolution issues without expense, the workflow turns to Upscale.media: - Sign up for an account under the free tier to unlock bulk image processing. - Navigate to the upload area, add all generated character stills, and view them under **Storage**. - Select all images, click **Transform**, and navigate to the upscale controls. - Set the scale factor to **2x**, which the creator notes is plenty for final output clarity. - Export and download the processed stills bundled into a single ZIP archive.

[03:13](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=193s) Audio Sourcing and Voice Cloning with ElevenLabs

Once visual stills are secured, production shifts to generating dialogue. The creator chooses Harry Potter as the initial speaking subject and heads to YouTube to find reference audio, entering the search query `Daniel Radcliffe voice for 1 minute`. The goal is to locate a clean, high-fidelity sample of the actor's natural speaking voice without background music or heavy distortion.

The selected YouTube URL is pasted into an online YouTube-to-MP3 converter to extract the raw audio track. Because long files are cumbersome and may include unwanted artifacts, the file is routed through an online MP3 cutter tool to isolate the cleanest section of clean speech. The isolated sample is then brought into ElevenLabs: - Open ElevenLabs and select the **VoiceLab** tab. - Choose **Add Voice** and select **Instant Voice Cloning**. - Upload the trimmed MP3 sample, name the profile, and click **Add Voice**. - Use the newly cloned vocal model inside the speech synthesizer to generate the intended fashion-spoof script lines. - Download the generated speech audio files and repeat the sample-and-clone sequence for any other character scheduled to speak.

[04:02](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=242s) Animating Characters with D-ID

To animate the static portraits into subtle moving figures and lip-synced speakers, the video relies on D-ID, which grants new users 20 free trial credits. The creator points out that every video requires two distinct types of animations: characters that speak their lines aloud, and ambient, non-speaking runway poses to cut between during the edit.

For speaking characters: - Select **Create Video** and click **Add** to upload the upscaled character portrait. - Switch the panel toggle from **Script** over to **Audio**. - Upload the cloned ElevenLabs speech file, click **Generate Video**, and download the resulting MP3-synced visual once processing finishes.

For silent runway reactions: - Upload the character image into a fresh D-ID project canvas, including figures who also have speaking lines so you have alternative reaction footage. - Leave the toggle set to **Script**. - Instead of typing text, click the small clock icon, which injects an artificial pause duration of 0.5 seconds per click. - Click the icon six consecutive times to build an empty 3-second block of ambient motion: “no need to do anything else, just hit generate video.” - Download all silent animated portrait files once rendered.

For the soundtrack, the tutorial suggests the classic track popular in other viral Balenciaga spoofs ("Lightvessel" by Thip Trong). The creator warns that creators hoping to monetize must obtain licensing permissions via Epidemic Sound or choose alternative royalty-free cuts.

[05:16](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=316s) Final Editing and Camera Effects in CapCut

The final assembly takes place inside CapCut, chosen because it is free, supports browser and mobile platforms, and handles rhythmic fashion-show transitions smoothly. The creator personally prefers the mobile interface for pacing and timeline precision.

The editing sequence unfolds as follows: - Import all speaking clips and 3-second ambient loops into the project timeline, arranging their playback sequence to match the pacing of the runway show. - Layer the background music track across the timeline beneath the video clips. - Generate runway flash photography effects by splitting the video track at major rhythmic beats, tapping **Transition**, navigating to the **Light** category, and choosing **Flashes**. - Go to the **Audio** menu, open **Effects**, locate the **Camera Shutter 2** sound effect, and align it precisely beneath every flash transition cut. - Inject continuous movement into the AI clips: “Make sure to have a dynamic zoom in every clip as it makes the scene more alive.” - Apply an initial keyframe at the start of a clip at normal framing, place a second keyframe at the end of the clip, and scale the image up slightly to create a smooth, subtle push-in throughout the shot.

Wrapping up the walkthrough, the creator invites viewers to share any alternate tools, shortcuts, or workflow optimizations in the comments below, reminding the audience to like and subscribe before closing out the tutorial.

## Notable quotes > [05:16](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=316s) Make sure to have a dynamic zoom in every clip as it makes the scene more alive. — PromptJungle

> [01:53](https://www.youtube.com/watch?v=TGD8zKvRxc4&t=113s) Always envision the video while you are choosing your images for the best outcome and a higher quality content. — PromptJungle

## Mentioned PromptJungle, ChatGPT, Midjourney, Upscale.media, ElevenLabs, D-ID, CapCut, Epidemic Sound, YouTube, Harry Potter, Balenciaga, Discord, Synthesia, HeyGen, Colossyan, Thip Trong, Lightvessel, Dan Henig, Danger Snow

## Video description 🔎 Have a question or an issue? 🎉 Join our community on Discord for in-depth discussions, tips on AI tools, and more! 🚀 Here's your invitation: https://discord.gg/v9zRyn8smW Looking forward to seeing you there! 👋

Alternative to D-ID. Try https://bit.ly/412eNx0 😊 And here is a comparison of the two tools: https://www.youtube.com/watch?v=2gqHaOkkQ5c

🔹 Music track for the video - Thip Trong - Lightvessel Get 30 days for free and the copyright for it with epidemicsound if you gonna use it for monitization on youtube. https://share.epidemicsound.com/fez7yf/?playlist=fguxfkbxipdhm3sh8hyu9ogtlf5vnezo

Here is How to make Harry Potter by Balenciaga Video: The actual video - https://www.youtube.com/watch?v=_N8RX3lIQDU

Lets start:

🔹 Generate the image prompts with ChatGPT

ChatGPT - https://chat.openai.com/chat ChatGPT prompts:

1. "Give me the 10 most popular Harry Potter characters, just the names."

2. "Now, act as a Balenciaga fashion designer with 20 years of experience. I want you to dress each of these 10 characters in a unique Balenciaga piece for a runway show. Just write the name of the character and the piece. Remember, it's a 1990 fashion show."

3. "Ok, now this is a prompt that I use to generate images from another AI image generation tool called Midjourney: '1990 screengrab of female model Hermione Granger wearing a grotesque black Balenciaga dress, fashion movie scene, Balenciaga commercial --ar 3:2 --v 5'. Just say 'Y' if you understand."

4. "I want you to rewrite the prompt but for every character, with the clothes you chose for them. Remember to write the prompt exactly as I pasted it, with the same format and parameters. Only change the gender of the model, the name of the character, and the description of the clothing. Do you understand? Please ask if you need more information."

If you are very new to ChatGPT this 2 min video will help you get started. https://www.youtube.com/watch?v=AXn2XVLf7d0

🔹 Generate the images with MidJourney MidJourney (click join the beta) - https://www.midjourney.com/

If you are new to MidJourney this 5min tutorial is a great start. https://www.youtube.com/watch?v=90tXPbyRGS8

🔹 Upscale your images if they have poor resolution with Upscale.media Upscale.media - https://www.upscale.media/

🔹 Generate the voiceovers with Elevenlabs. Elevenlabs - https://beta.elevenlabs.io/

If you are lost in the cloning part watch this 1min video. https://www.youtube.com/watch?v=3AtAiGtPi4s

Youtube to MP3 converter - https://en.onlymp3.to/164/ MP3 cutter - https://mp3cut.net/

🔹 Generate the AI animation with D-ID D-ID - https://www.d-id.com/

If you want to use an alternative to D-ID. Try https://bit.ly/412eNx0 😊 And here is a tutorial on how to use it. https://www.youtube.com/watch?v=2gqHaOkkQ5c

🔹 Music track for the video - Thip Trong - Lightvessel Get the copyright for it with epidemicsound if you gonna use it for monitization on youtube. https://share.epidemicsound.com/fez7yf/?playlist=fguxfkbxipdhm3sh8hyu9ogtlf5vnezo

Altenative copyright free song : (follow artist's terms&con) Dan Henig | Danger Snow https://www.youtube.com/watch?v=X4KYTFEJuFI&t=0s

🔹 Edit with capcut https://www.capcut.com/

In this step-by-step tutorial, learn how to create AI animated Harry Potter characters in the meme Balenciaga style. I will walk you through the entire process, from generating images using AI tools like ChatGPT and text-to-image tools like Midjourney, to creating immersive voiceovers with ElevenLabs, and finally, animating your characters with D-ID's text-to-video AI technology. And I also show you how to put all togother at the end.

‐------------------

- BACKGROUND MUSIC OF THIS VIDEO https://share.epidemicsound.com/fez7yf/?playlist=au0ut0ba17wivw3ipbbg1f9abaq4cba5

- VOICEOVER OF THIS VIDEO Synthesia - https://bit.ly/3yKHmDa

-------------------

🤖 OTHER AI TOOLS that I HAVE REVIEWED before*

Synthesia - https://bit.ly/3yKHmDa HeyGen (Movio)- https://bit.ly/412eNx0 Colossyan Studio: https://bit.ly/3M6kKoC Elai - https://bit.ly/400pJvb D-ID - https://bit.ly/D-IDPromptJungle DEEPBRAIN - https://bit.ly/HumanAIPromptJungle

My Other AI Animated Avatars videos:

How To Create a Talking AI Avatar for FREE (better than synthesia) https://www.youtube.com/watch?v=kYueHEcqdmM

Top 5 AI Generated Talking Avatar Tools (D-ID Alternatives) https://www.youtube.com/watch?v=U0HNdsxC8YU

If my work speaks to you, hit the share button and spread the love on your socials. And if you're feeling extra kind, a coffee from you would be the cherry on top of my creative sundae! 🎶🌞☕️ https://linktr.ee/promptjungle *The AI links above are PromptJungle's affiliate links. If you make a purchase with my link there is no additional cost to you and in return PromptJungle makes a small commission. #tutorial #ai #harrypotter #balenciaga

[Source](https://www.youtube.com/watch?v=TGD8zKvRxc4)

中文简报

巴黎世家风《哈利·波特》分步制作教程 该视频概述了制作“巴黎世家风《哈利·波特》”时尚恶搞视频的端到端工作流程，无需付费课程或复杂设置。观众将学习如何使用 ChatGPT 设计格式化的 Midjourney 提示词，通过 Upscale.media 放大生成的肖像，并在 ElevenLabs 中利用 YouTube 音频样本克隆演员声音。随后，教程演示了如何在 D-ID 中生成说话动画和无声走秀循环片段，最后在剪映（CapCut）中结合动态缩放、相机闪光灯转场以及快门音效完成最终剪辑。 巴黎世家风格的 AI 梗图制作流程依赖于整合多种工具的流水线：使用 ChatGPT 编写提示词、Midjourney 制作视觉效果、ElevenLabs 进行声音克隆，以及 D-ID 生成动画。 (0:00) 通过使用 --ar 3:2 和 --v 5 等特定参数可保持 Midjourney 的图像质量，而在提示词中加入“grotesque”（怪诞）等主题关键词则能维持美学风格的一致性。 (1:51) D-ID 动画任务分为两类：用于对白口型同步的音频文件，以及通过手动插入停顿生成的 3 秒无声循环片段（用作走秀 B-roll 画面）。 (4:02) 在剪映中进行最终润色时，需要卡点添加“相机快门”音效和“闪光灯”转场以模拟秀场摄影体验，并配合慢速缩放关键帧来让静态 AI 肖像动起来。 (5:16)

Reading summary

PromptJungle outlines an end-to-end, free pipeline for creating viral Balenciaga-style AI fashion memes, combining ChatGPT, Midjourney, ElevenLabs, D-ID, and CapCut to generate and assemble character animations. Text Prompt Structuring and Formatting via ChatGPT (0:00) The workflow begins by leveraging ChatGPT in a multi-step prompting sequence to generate consistent runway themes and formats compatible with image generators. The user prompts the AI to act as a 1990s Balenciaga fashion designer with 20 years of experience dressing ten selected Harry Potter characters. Once the runway pieces are paired with character names, the creator trains ChatGPT on a specific Midjourney syntax template: '1990 screengrab of female model Hermione Granger wearing a grotesque black Balenciaga dress, fashion movie scene, Balenciaga commercial --ar 3:2 --v 5'. The model then outputs formatted prompts for all ten figures, preserving parameters while altering only gender, character name, and runway attire. Visual Generation, Stylistic Corrections, and Batch Upscaling (1:51) Prompts are fed into Midjourney's Discord channels to render base character stills. To manage facial consistency and styling drift, the creator advises removing phrasing such as 'male or female model' if generations fail to resemble the original actors, or re-prompting ChatGPT with targeted aesthetic descriptors like 'goth', 'grotesque', or constrained black-and-white palettes. After selecting the strongest generations, the images are imported into Upscale.media under its free tier for bulk processing. Setting the scaling factor to 2x provides sufficient detail and resolution for video production without added expense. Voice Cloning with ElevenLabs and D-ID Animation (3:13) Audio dialogue is constructed by sourcing clean reference speech from YouTube, such as Daniel Radcliffe's voice without background music, which is trimmed using an online MP3 cutter. ElevenLabs' Instant Voice Cloning under VoiceLab ingests the clean sample to synthesize scripted fashion spoof lines. To animate the portraits, D-ID produces two separate asset types: speaking sequences generated from uploaded ElevenLabs audio files, and ambient runway poses. The creator builds silent 3-second animations by stacking six 0.5-second pause blocks using D-ID's clock icon in the script panel, generating subtle movement for reaction cuts without spoken lines. Timeline Assembly, Flash FX, and Keyframing in CapCut (5:16) The final video is constructed in CapCut by sequencing spoken clips with ambient loops alongside a backing track, such as 'Lightvessel' by Thip Trong (noting that commercial monetization requires licensing via Epidemic Sound or alternative royalty-free cuts). To evoke runway fashion shows, the editor splits video tracks on musical beats to insert 'Flashes' transitions from the Light panel, synchronized directly with 'Camera Shutter 2' audio effects. Dynamic keyframing—scaling slightly from an initial keyframe to a final keyframe—is applied to each AI clip to produce subtle push-in zooms and prevent static visual presentation. The presented workflow demonstrates how free AI and video editing tools can replicate high-engagement viral animations, though users seeking to monetize output must manage commercial music licensing and account for credit limits across services like D-ID.

精炼阅读摘要

PromptJungle 完整演示了制作流行于社交平台的“巴黎世家（Balenciaga）”风格 AI 动画的免费工作流，并以《哈利·波特》模因为例，展示了从提示词设计到最终剪辑装配的六步管线。 ChatGPT 提示词工程与角色设定 (0:00) 制作流程首先在 ChatGPT 中构建结构化提示词。创作者先要求 ChatGPT 列出 10 位最受欢迎的《哈利·波特》角色姓名，随后赋予其“拥有 20 年经验的巴黎世家时尚设计师”角色设定，指示其为这 10 位角色分别设计适合 1990 年代时装秀的专属走秀服饰，且仅输出角色名与服饰描述。 为使生成的服饰描述适配 Midjourney 的绘图语法，创作者向 ChatGPT 提供了一套母版模板提示词（'1990 screengrab of female model Hermione Granger wearing a grotesque black Balenciaga dress, fashion movie scene, Balenciaga commercial --ar 3:2 --v 5'）。在确认 AI 理解该规范后，创作者指示其为全部 10 个角色重写该提示词，要求严格保留画面比例与版本参数（--ar 3:2 --v 5），仅替换性别、角色姓名与服装描述。 Midjourney 生成与批量超分辨率放大 (1:51) 在 Discord 的 Midjourney 新手频道中，创作者通过 /imagine 命令批量生成角色肖像。针对生成人物面部与原角色特征不匹配的问题，创作者建议可直接从提示词中删去“male model”或“female model”；若服饰氛围偏离预期，则可返回 ChatGPT 补充“哥特（goth）”、“怪诞（grotesque）”或限定“仅黑白（only black and white）”等修饰词。创作者特别指出，挑选图片时需提前构思最终成片的视觉连贯性。 在 Discord 中完成基础放大后，若图像仍缺乏制作视频所需的细节，工作流采用 Upscale.media 的免费层级进行批量处理。用户在平台注册后上传所选肖像，在“存储”面板全选图像并执行“变换”，将放大倍数设为 2 倍（创作者认为 2x 放大足以满足成片清晰度需求），最后将处理好的高清素材统一打包下载为 ZIP 压缩文件。 语音克隆与 D-ID 双轨动态资产制作 (3:13) 音频制作阶段以哈利·波特作为首个说话主体。创作者在 YouTube 搜索无背景音乐干扰的高保真原声片段（如检索“Daniel Radcliffe voice for 1 minute”），经由在线 YouTube 转 MP3 工具及音频剪辑网站截取最纯净的人声。随后在 ElevenLabs 的 VoiceLab 中使用“即时声音克隆（Instant Voice Cloning）”建立声音模型，并输入戏仿台词生成配音音频。 动态人物渲染依托 D-ID 平台（新用户拥有 20 次免费点数）。创作者强调成片需要两种镜头资产：一种是说话镜头，通过上传高清肖像并切换至音频面板，导入 ElevenLabs 配音生成对口型视频；另一种是秀场静态反应镜头。制作无声镜头时，用户需保持脚本面板开启且不输入文本，仅需连续点击时钟图标 6 次（每次添加 0.5 秒停顿），即可合成 3 秒的环境微动素材。对于背景音乐，教程推荐模因常用的《Lightvessel》，但创作者提醒若计划变现，须通过 Epidemic Sound 取得商业授权或选用其他免版税音乐。 CapCut 剪辑装配、快门闪光与推镜 (5:16) 视频的最终剪辑在免费工具 CapCut 中完成，创作者因节奏与时间轴控制精度偏好使用其移动端应用。首先将说话片段与 3 秒无声环境镜头按走秀节奏交替排列在时间线上，并在下方铺设背景音乐。 为还原秀场摄影氛围，剪辑师需在音乐重音处分割视频轨道，添加“光效（Light）”类别下的“闪光（Flashes）”转场，并在音频特效库中调用“相机快门 2（Camera Shutter 2）”音效，使其与每一次闪光精确重合。此外，创作者强调必须为每个剪辑片段添加动态变焦以赋予画面生机：在片段起始处添加初始关键帧，在结尾处添加第二帧并将画面适度放大，实现平滑缓慢的推镜动效。 该方案展示了如何组合免费或试用阶段的 AI 工具链完成高品质视觉生成，但涉及商用变现时仍受限于背景音乐的版权许可，创作者亦鼓励社区探索替代工具以进一步精简流程。

Source quotes

[5:16] PromptJungle: Make sure to have a dynamic zoom in every clip as it makes the scene more alive. [1:53] PromptJungle: Always envision the video while you are choosing your images for the best outcome and a higher quality content.

Original transcript

[0:00] so let me show you how to make these viral  Balenciaga AI animation videos many creators [0:06] charge for the prompts but I'll share everything  for free all I ask is for you to like this video [0:12] so it can reach more people as it really helps  the channel I'm going to link everything down [0:17] in the description and I will also link Sun short  beginner's guides for the tools if it's your first [0:22] time using them I'm going to use the Harry Potter  by Balenciaga video as an example in this tutorial [0:29] okay let's start to generate the images we will  use chat GPT and mid-journey I will share the [0:37] main prompts in the description as a list so  it's easier for you to copy and paste first go [0:43] to chatgpt open a new chat and type give me the  10 most popular Harry Potter characters just the [0:50] names after that paste this prompt now act as  a valencyogger fashion designer with 20 years [0:58] of experience I want you to dress each of these  10 characters in a unique Balenciaga piece for a [1:03] runway show just write the name of the character  and the piece remember it's a 1990 fashion show [1:11] when it has finished matching the clothes with the  characters we need to give it an example of our [1:16] main prompt so paste the next prompt you have in  the list okay now this is a prompt that I use to [1:23] generate images from another AI image generation  tool called mid-journey 1990 screen grab a female [1:30] model Hermione Granger wearing a grotesque black  belt seaga dress fashion movie scene Balenciaga [1:37] commercial R3 colon 2v5 just say yes if you  understand after that paste the next prompt [1:45] which will tell it to generate 10 prompts for  your images that you will generate in mid-journey [1:51] now add to Discord to use mid-journey log in  and go to the rookie Channel and start feeding [1:57] those prompts one by one tight slash imagine and  paste your prompts choose the ones you like and [2:05] upscale them if you start getting images that  don't resemble the character's facial features [2:10] you can always delete the part where it says  male or female model also if you start getting [2:16] clothes that don't match the theme of the video  you can go back to chat GPT and instruct it to [2:22] rewrite the prompts with a theme in mind like goth  grotesque or even colors like only black and white [2:29] go through all your images and download  the ones you like always Envision the [2:34] video while you are choosing your images for  the best outcome and a higher quality content [2:39] if you're not happy with the quality of the images  head to upscale.media to get a better resolution [2:45] for your images click on upload image and sign  up so you can bulk upload choose the free plan [2:52] and upload all your images then click storage  select all your images and click transform [2:59] scroll down and click upscale choose your scale  I'm going to choose times 2 as I think that's [3:06] enough and click export into a zip file download  your images when done now let's generate the audio [3:13] part choose the characters that will talk in  your video I'm going to do Harry Potter so [3:19] go to YouTube and search for Daniel Radcliffe  voice for one minute find a good quality sound [3:25] in the video and copy the link head to YouTube  to MP3 converter and convert and download the [3:31] MP3 file if the file is too long or you want  to isolate a part of the file head to an MP3 [3:38] cutter and cut and download the part you want to  use after that go to 11 labs and click voice lab [3:46] then add voice instant voice cloning and upload  the MP3 file of your character click add voice [3:56] then use it to generate your script and download  the generated voiceover do this with all the [4:02] characters that will speak in your video now for  the animation part head to did and sign up you [4:09] have 20 credits for free click create video then  add and choose your first character we will do two [4:17] different animations the first one is animating  the speaking characters so toggle to audio [4:24] upload the voice of the file and click generate  video download the video when ready do this with [4:31] all your talking characters after that we need  to animate all the other characters in our video [4:37] both talking and Silent this time toggle to  script instead and click on the clock icon [4:43] which indicates a pause of 0.5 seconds so I will  choose 6 pauses which totals 3 seconds no need to [4:52] do anything else just hit generate video do this  with all your characters when done download all [5:00] the clips for music if you want to use the same  legendary background track that everyone uses [5:06] I've linked it in the description but remember  that you will need consent to use it on your [5:11] YouTube channel you can find the copyright for  it on epidemic sounds which I'll link as well [5:16] now to put the whole video together you can use  any software you're comfortable with I'm going to [5:22] show you how to do it with cap cut because  it's free and you can do the camera flash [5:27] effects easily you can use it on your browser  or your phone I'm more comfortable using it on [5:33] the phone so open all your clips and reorder them  as you wish add your background track and to make [5:40] the flash effect choose the place on your timeline  and split then tap on transition and choose light [5:47] then flashes after that go to audio then effects  and find the camera shutter 2 effect align it [5:56] under every flash transition make sure to have a  dynamic zoom in every clip as it makes the scene [6:02] more alive to do that just put one keyframe at the  start of the clip and another at the end of the [6:08] clip and have an end Zoom position of the clip and  you're done this is how it will look like if you [6:15] know any other shortcuts or ways that are better  or easier to do any part of this tutorial please [6:20] don't hesitate to share it in the comment below  with the others and if you found this helpful [6:25] don't forget to like the video and maybe subscribe  as always see you in the next one foreign