← All transcripts

You Can Finally BRIEF an AI Video Model — I Fed It 50 References Transcript, AI Summary & Key Points

ManuAGI - AutoGPT Tutorials · 11 hours ago · Science & Technology · 10:52 · EN-US

💡 Answer

Yes. Seedance 2.5 can be briefed with up to 50 references in one generation, allowing creators to provide characters, locations, looks, movement, and audio instead of relying only on a text paragraph.

🧠 AI Summary

Seedance 2.5 is presented as an AI video model that accepts up to 50 references in one generation: 30 images, 10 video clips, and 10 audio files. It supports motion, creative, and clay-render references, videos up to 30 seconds with multiple shots, extensions that carry over characters and pacing, and multi-character consistency. The demonstrated workflow generates a story idea and master prompt, creates a character storyboard image, then uses that image and the video prompt to generate a 30-second 3D animation. The main conclusion is that reference-based direction gives creators more control over characters, voices, style, movement, and scenes than text-only prompting.

🔑 Key Points

  • Seedance 2.5 supports up to 50 references in one generation: 30 images, 10 video clips, and 10 audio files.
  • Motion references communicate desired movement through footage, while creative references preserve mood and look across a video.
  • Clay-render references can provide camera paths, positioning, movement, and lighting direction for a finished scene.
  • Seedance 2.5 can produce up to 30 seconds in one pass, include multiple shots, and extend an existing video while carrying over characters and pacing.
  • In group scenes, characters are intended to retain their individual faces and voices.
  • The demonstrated workflow produced a 30-second 3D animation from a character storyboard image and a detailed video prompt.
  • The presenter concludes that the primary benefit is increased control over AI video rather than simply generating a one-take clip.

✅ Actionable items

  • Use motion footage as a reference when directing movement instead of describing the movement only in text.
  • Use creative references to establish the intended mood and visual look.
  • Use a clay-render reference to communicate camera paths, character positions, movement, and lighting direction.
  • Create a story idea first, then add the selected story title to the master video prompt.
  • Generate separate character prompts when creating characters for an animation.
  • Check that uploaded media is eligible before using it in a generation.
  • Verify that you own the rights to the content or have permission to use it before viewing or downloading the result.
  • Brief, generate, adjust, and repeat to improve results by volume.
  • Use the platform on desktop for the demonstrated workflow.

💡 Business ideas

AI-assisted production of 3D cartoon short videos00:53

Create a 3D animation story by developing a story concept, generating character and storyboard reference images, then using a multi-reference AI video model to produce a scene with consistent characters, voices, locations, style, motion, and audio.

For
Creators who want to make 3D animation stories or AI-generated video without manually producing every visual element.
Solves
Text-only AI video prompts make it difficult to communicate the intended characters, locations, lighting, music, movement, and visual style, while character faces and voices often lose consistency across group scenes.
  • The Little Clock Maker: generated a storyboard character image with 11 image panels and then a 30-second animated video; the creator described the final result as really amazing.
Soon you can unlock the full business plan.

Behind this: 13 build steps · 4 tools and how each is used · how to validate demand · 3 more real examples · 6 things the video never answers.

Inquire for details

Soon you can unlock the full business plan.

Inquire for details

🤖 AI in practice

Used for

Generate five creative 3D short-animation story ideas. 03:00
Turn a selected story title into a complete Seedance 2.5 video-generation prompt. 04:12
Create detailed prompts for each character in an animation story. 05:05
Generate storyboard character images for an animated story. 05:36
Generate a multi-shot animated video from a storyboard image, a detailed prompt, and visual, motion, creative, and audio references. 01:48

Advice

  • Use ChatGPT to generate multiple story ideas when you do not already have one. for Animation creators and AI video makers
    It can produce five original 3D cartoon short-animation ideas in a few seconds.
  • When using the Seedance 2.5 master prompt, change only the story-title section and leave the rest unchanged. for Seedance 2.5 users
    The completed prompt is designed to produce the full story video prompt for Seedance 2.5.
  • Use references to show motion, mood, visual style, camera paths, character positions, and lighting instead of describing everything in text. for AI video creators and directors
    Motion references communicate movement, creative references preserve look and mood, and clay-render references provide scene and camera guidance.
  • Generate repeatedly: brief the model, review the result, adjust, and generate again. for Seedance 2.5 users
    The creator says the model improves through volume.
  • Use Higgsfield on desktop for this workflow. for Higgsfield users
    The creator says the workflow runs best there.
  • Only confirm rights verification when you own the content or have permission to use it. for AI video creators using reference content
    The platform may request rights verification before the generated result can be viewed or downloaded.

What it could not do

  • AI video can normally lose character identity and voice consistency in group scenes. — The creator presents multi-character group scenes as a point where AI video usually falls apart, although Seedance 2.5 was reported to preserve each person's face and voice in this test.

🧰 Tools & AI usage

  • Higgsfield AI — Provides the image and video generation platform used to create character references and generate the final animation.01:12
  • ChatGPT — Generates story ideas, complete Seedance video prompts, and character prompts.03:00
  • GPT Image 2 — Generates the character or storyboard reference image from a detailed character prompt.05:45
  • Seedance 2.5 — Generates up to 30-second multi-shot video scenes from prompts and image, video, and audio references while maintaining character and voice consistency.07:27

AI is used for

  • Generate story ideas — Produce five creative 3D short-animation concepts when the creator does not already have a story idea.03:00
  • Create a master video prompt — Turn a selected story title into a complete story video prompt for Seedance 2.5.04:16
  • Create character prompts — Generate detailed prompts for each character, including character details, panels, clothing, and other elements.05:16
  • Generate a character storyboard image — Create a visual character reference containing the story title, 11 image panels, character details, and prop details.06:28
  • Generate an AI video — Turn the storyboard character image and detailed video prompt into a 30-second 3D animation.07:17

📄 Transcript

Searchable transcript of You Can Finally BRIEF an AI Video Model — I Fed It 50 References — ManuAGI - AutoGPT Tutorials (10:52). Search for a phrase, then click its timestamp to jump straight to that moment in the video.

Captions sourced from the original video on YouTube, published by ManuAGI - AutoGPT Tutorials. The video, its captions and all related intellectual property remain the property of their respective owners; AINotes claims no ownership. Provided for research, accessibility and search — see the Transcript Notice and Copyright Policy.

00:00 Every AI video tool has the same problem. You type a paragraph, cross your fingers, and hope it understood you. You can't show it what you mean. You can't hand it your character, your location, your lighting, your music, and say, "Make it like this." Until now. I just handed an AI 50 references in a single generation. 30 images, 10 video clips, 10 audio files, my whole cast, my location, my look, my track, and this is what came back.

00:33 30 seconds, one generation, every face held, every voice held. Cedense 2.5 is here. And this is the first time it feels like briefing a crew instead of rolling dice. I'll show you exactly how it works live and how to get 33 days of unlimited access. And I will show you how to make 3D cartoon videos like this. First, real quick, if you're new here, welcome to Manu AGI.

00:58 I'm Manu and on this channel, I show you the best AI tools and how to use them in simple steps anyone can follow. If that sounds helpful, hit subscribe. Okay, let's make a AI animation story. So this is a Higsfield AI all-in-one AI powerful platform. Access the world's top AI models, tools, and capabilities all in one place designed to help you create, innovate, and get more done.

01:23 And Cedence 2.5 upgrades three things and everything follows from them. First, how much you can hand it. Up to 50 references in one generation. 30 images, 10 video clips, 10 audio files. And there are new kinds of references that matter more than the number. Motion references. Show it footage of the movement you want instead of writing a paragraph about it.

01:48 Creative references. Show it the mood and look and it holds that direction across the whole video. And clay render references. Hand it a plain gray 3D mockup and it reads your camera path where people stand, how they move, even your lighting direction, then builds the real scene on top of it. Hello. >> Second, how long the story runs. Up to 30 seconds in one pass.

02:33 Double 2.0. And it's a scene, not one long moment. Multiple shots can live inside it. And you can extend it, adding the next shot onto a video you already made with characters and pacing carrying over. Third, multicar consistency. >> Birthday buddy. I did all of this alone. >> In group scenes, everyone keeps their own face and their own voice. That's exactly where AI video normally falls apart.

03:14 So, let's test it. I'm going to load it up and see how much of my direction actually survives. First, go to chat GPT or Claude to create your story idea prompt. Before creating your story, we first need a story idea. If you already have a story idea, that is perfect. But if you do not have one yet, you can use this prompt to generate new ideas for your story.

03:36 For example, I want to create a new 3D animation story. So, I will enter this prompt. I want to create a 3D short animation story. Give me five creative and engaging story ideas for my story. Now, press enter. In a few seconds, Chat GPT will create five original and 3D cartoon short animation story ideas. For example, the moon's missing star, the robot who wanted to dream, the dragon who was afraid of fire, and two more story ideas.

04:06 You can choose any story idea you like. If you do not like the ideas, you can generate more ideas. Again, after selecting your story idea, copy my master prompt for Cedense 2.5. This master prompt is also available in the video description. So, do not worry. Just copy the complete master prompt. Now go back to chat GPT and paste it into a new chat. At the top of this prompt, you will see a section for the story title.

04:36 This is where you need to enter your selected story title. For this video, I am choosing this story idea, The Little Clock Maker. So, I will copy this title and paste it into the story title section. Do not change anything else in the master prompt. Just add your story title and press enter. In a few seconds, chat GPT will create your complete story video prompt for Cance 2.5.

05:02 Now, our master video prompt is ready. But I need a my own character for this video right now. So, if you'd like to include your own character, you can. Or I want to generate a character based on my video's theme. So, I'll simply type it in here. Create my character prompts for the little clock maker story. So, we create my each character separate. Now, chat GPT will create a full prompt for your character based on your video script, including all the details of the character, all the panels, the way the character is

05:31 worn, and everything else. Now, you just need to copy this prompt and simply go on Higsfield AI. On Higsfield, you will see a dashboard like this. From the top menu, go to the image tab and choose the GPT image 2 model. Now, for image creation, you will see a dashboard like this. If you have any reference images, you can upload them here. You can upload up to 16 images.

05:56 But for this example, we do not have any reference images, so we will only use our prompt. We already copied our character master prompt. So now we simply paste it here. Next, choose your model. We already selected GPT1 mage 2, but if it is not selected on your side, then choose it here. After that, choose the image quality as high. Next, choose the image resolution.

06:20 It supports up to 4K. So, for this image, I will select 4K. Now, choose the image aspect ratio. By default, it is set to auto. But for this image, I want 16 to9. So, I will select the 16-9 aspect ratio. Next, choose how many images you want to generate. For example, I want two images. So, I select two. After selecting everything, I click on the generate button.

06:46 And after a few seconds, my two images are ready. You can see them here. The storyboard character includes the story title, the little clock maker, along with 11 image panels. And you can see all the character details along with other important elements and prop details. Once you are satisfied with your story character image, just click on the download button to save it.

07:09 Or if you want to create it again, simply click on the recreate button. So we download this image. Now after downloading and preparing your story character image, the next step is to turn this image into a video according to our master video prompt using Cedense 2.5. For this imagetovideo process, we need a full detailed prompt for Cedense. So now we go back to chat GPT.

07:34 And as you can see here, we simply copy Cedence 2.5, the little clockmaker story master prompt, and go back to Higsfield AI. Here, go to the video section and select the seedance 2.5 model. You will see a dashboard like this. First, upload your storyboard character image. You can upload it using the upload media button. Up to 50 references in one generation, 30 images, 10 video clips, 10 audio files.

08:04 Also, one useful feature in Higsfield AI is the check eligibility option. It checks whether your image is eligible to use on the platform. After a few seconds, my image is approved. So, I select it. Next, paste your seedance 2.5 video prompt here. Next, choose your model. We already selected seedance 2.5, but if it is not selected on your side, choose it here.

08:29 Now, choose your video duration. The default duration is 8 seconds, but for this video, I need a longer result, so I select 30 seconds. Next, choose the video aspect ratio. For this example, I select 16-9. Then, choose your video resolution, 1,080p or 720p. Here, I'm using Cance 2.5 unlimited. You can also access it, which is I already explained in this video.

08:57 After selecting everything, click on the generate button and the video generation process will begin. Before viewing or downloading the result, it may ask for rights verification. I own rights to this content only if you truly own the content or have permission to use it. Now my final result is ready. Let's watch the result. We have to fix it. Talk.

09:31 I can't fix it. Tick, tick, tick. Tomorrow we fix that. >> It looks really amazing. Higsfield created the full video. Honest verdict. The 30-second one take gets the headlines. But the real story is control. 50 references going in, secondby-second editing coming out, and multi character scenes that actually hold. For the first time, you're not fixing AI video, you're directing it.

10:01 And right now you can get 33 days of unlimited cedence on Higsfield, which is exactly what a model like this needs because you get good by volume. Brief it, generate, adjust, go again. Links in the description. Quick tip for the best experience. Use Higsfield on desktop. That's where everything we just did runs best. That's Cedance 2.5. 50 references, 30 second scenes, and editing by the second.

10:29 33 days of unlimited access at the link below. If this was useful, subscribe. I'm pushing this model a lot further on this channel. And tell me in the comments, what would you put in your 50 references? Best answer gets pinned. Click the video on screen to keep going. See you in the next