Big Idea

The argument in one line.

A three-tool chain, a chat model to write the prompt, an image generator to render the still, and Seedance 2.5 to animate it, turns motion graphic work that took hours of manual keyframing into a few minutes of prompting, provided the animation prompt explicitly protects the on-screen text.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You edit or design motion graphics and want to see whether an AI video model can now handle the animation work you'd normally keyframe by hand.
You want the exact prompt chain (chat model deconstructs an image, image model renders it, video model animates it) rather than just a feature announcement.
You want to know what's specifically new in this model update: longer generations, more reference images per prompt, and text that survives animation intact.

SKIP IF…

You're looking for color grading, sound design, or timeline-editing techniques, not AI image-to-video generation.
You need a tool that works without an internet connection or outside a paid generation credit system.

TL;DR

The full version, fast.

Higgs Field's Seedance 2.5 update generates up to 30 seconds of video from one prompt, accepts up to 50 reference images, and fixed a longstanding problem where on-screen text warped during animation. The demonstrated workflow chains three tools: a chat AI (Gemini, Grok, or Claude) deconstructs a reference image and writes a detailed generation prompt, an image model (GPT Image 2 or Nano Banana Pro) renders the still at 4K, and Seedance 2.5 animates it from a short motion instruction that explicitly asks to keep the text intact. A build that used to take about 20 minutes of manual editing took roughly 3 minutes end to end. When no single reference fits, screenshotting several unrelated images and describing the combined animation to the chat model works too.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 00:24

01 · Cold open: the claim

States the thesis up front: reference images animate into finished motion graphics with almost no errors, cutting work that used to take hours of hand animation and years of practice down to a few minutes.

00:24 – 01:39

02 · What's new in Seedance 2.5

Introduces the just-released Seedance 2.5 model inside Higgs Field: generations up to 30 seconds long, up to 50 reference images per prompt, and on-screen text that no longer warps during animation. Notes that Gemini and Claude are used to draft the generation prompts.

01:39 – 02:36

03 · Two workflows, then gathering references

Explains the two paths to a motion graphic: animate a chosen reference image directly, or have a chat AI write a fully detailed prompt with no image at all. Opens several Pinterest boards (Apple-style UI, cinematic shots, widgets, stylized graphics) to source starting images.

02:36 – 04:36

04 · Three builds: UI widget, connected widgets, house-label widget

Runs three examples back to back: an Apple-style UI widget animated with sliding bars and a button click (about 3 minutes versus an estimated 20 minutes by hand), a widget collage that pops in and connects with an animated string, and a labeled widget pair where the text writes out letter by letter.

04:36 – 05:41

05 · Cinematic chess piece and a multi-reference finale

Generates a cinematic gray-background environment from a chess-piece style reference, animates it with a rotating piece and bouncy text, then closes by screenshotting several unrelated Pinterest images and asking Gemini to describe a combined animation across all of them.

Atomic Insights

Lines worth screenshotting.

Seedance 2.5 inside Higgs Field can turn a single reference image into an animated motion graphic in about three minutes, work that previously took roughly 20 minutes by hand.
The update raised the generation ceiling to 30 seconds of video from one prompt and up to 50 reference images accepted per request.
The single biggest fix in this update is that on-screen text stays legible through the animation instead of warping the way earlier video models did.
There are two ways to build a prompt: hand the model a reference image and describe the motion, or have a chat AI write a fully detailed prompt with no image at all.
Asking a chat model to 'deconstruct this image and give me a prompt' produces a far more detailed and specific generation prompt than writing one from memory.
Changing one detail in a deconstructed prompt, a name or a button color, is enough to make a borrowed reference image look custom without redrawing anything.
Explicitly telling the model to 'keep all the text high-quality and intact, no blurriness' in the animation prompt is what prevents on-screen copy from garbling.
When no single reference fits an idea, screenshotting several unrelated images and letting the chat AI describe a combined animation gives the video model enough direction to invent a coherent scene.
Setting a short first-pass duration, four to six seconds, keeps drafts fast to review before committing credits to a longer render.
A cinematic environment (gray background, directional lighting, a floating object) can be generated from a single style-reference photo without an existing photo of the final scene.

Takeaway

The three-tool chain that turns a still into a motion graphic

AI Motion Graphics

Chaining a chat model, an image generator, and Seedance 2.5 replaces hours of manual keyframing with a few minutes of prompting, as long as the animation prompt explicitly protects the on-screen text.

Asking a chat model to 'deconstruct this image and give me a prompt' produces a far more detailed generation prompt than writing one from scratch.
Changing one detail in a deconstructed prompt, like a name or a button color, is enough to make a borrowed reference image feel custom without any manual editing.
An animation prompt that explicitly says 'keep all the text high-quality and intact, no blurriness' is what stops on-screen copy from warping during generation.
Setting a short duration, four to six seconds, on a first pass keeps drafts fast to review before spending credits on a longer final render.
When no single image matches the idea, screenshotting several unrelated references and describing the combined result to a chat model gives the video generator enough direction to invent a coherent scene.
A cinematic environment can be generated from one style reference and a plain description of lighting and background, without needing an existing photo of the final scene.
Naming the exact motion, a piece rotating, a bar sliding, a button click, produces more controlled results than a vague instruction to animate the image.

Glossary

Terms worth knowing.

Seedance 2.5: The AI video model used throughout the video to animate a still image into a short motion graphic, accessed inside Higgs Field's Cinema Studio interface.
Higgs Field: The AI creative platform (branded 'Cinema Studio') that hosts Seedance 2.5 alongside image generation tools, where prompts, reference images, and duration/resolution settings are set before generating.
GPT Image 2: An image-generation model used to render the static reference image from a detailed text prompt before that image is animated.
Nano Banana Pro: An alternative image-generation model offered inside Higgs Field alongside GPT Image 2 for producing the starting still image.
Deconstruct prompt: The technique of feeding a reference image to a chat AI and asking it to describe the image in detail as a reusable generation prompt, rather than writing the prompt from scratch.

Resources

Things they pointed at.

00:24toolHiggs Field (Cinema Studio) ↗

00:24toolSeedance 2.5

01:56toolGPT Image 2

04:00toolNano Banana Pro

00:53toolGemini

00:53toolClaude

00:59toolGrok

01:27toolPinterest

Quotables

Lines you could clip.

00:05

“Every image that I've thrown at it has been animated into perfect animations with zero problems or mistakes.”

the flat, testable claim that opens the video→ TikTok hook↗ Tweet quote

00:45

“The main thing I noticed with this update is it doesn't mess up text anymore.”

names the specific, checkable fix in one line→ IG reel cold open↗ Tweet quote

02:28

“It took about three minutes. And if I were to do this in an editing software, it would probably take me like 20 minutes.”

a concrete before/after time comparison→ newsletter pull-quote↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

I just didn't think it was going to be this good. So far, every image that I've thrown at it has been animated into perfect animations with zero problems or mistakes. I'm not exaggerating at all right now.

You can type in whatever you want, download that image, upload it into Cdance 2 .5, ask it to animate it, and what used to take five hours of work and five years of practice now takes five minutes. As an editor myself, I can't say that I like that, but I also don't not like it. So let me show you what I've been able to create and then you can try it yourself So we're gonna be trying out a lot of different motion graphics so that you can see what works best Seedance 2 .5 just dropped a couple of hours ago in Higgs field So that's what we're gonna be using the link for that is down below But basically as you can see we have a slider here that can go up to 30 seconds So now we can generate videos up to 30 seconds in one prompt and we can actually upload up to 50 references now But the main thing I noticed with this update is it doesn't mess up text anymore Especially if you ask it not to mess up the text with the results being a lot more accurate I've been using lots of Gemini and Claude to build my prompts for me.

This obviously allows me to get more complex. motion graphics. So there are two ways we can go about creating these.

We can find a reference and then generate an image with GPT -2 or Nanobanana, and then ask Cdance to animate it. This is for if you know exactly what you want and need it to look a certain way, because essentially it's just going to follow the image you gave it. Or we can ask an AI chat, like Grok or Gemini, to generate us a super in -depth prompt, and then we can just paste that in directly.

We're going to get to this part a little bit later, and I'm going to upload a bunch of references, and you're going to see how that works. So as you can see, I've opened up some Pinterest tabs here. Here's one for kind of Apple UI.

Here we have kind of a more cinematic approach. Here I found some widgets. And then for a more stylized motion graphic, I found this image.

So what I'm going to do is save all of these as an image. And then we can drag this image into an AI chat. For this case, I'm just using Gemini because it's really cheap.

And if we use the pro version here, I found it gives us really good prompts. I'm just going to write deconstruct this image and give me a prompt for a new image. I just want you to change the text and change the name of the button in the corner to Zane and also make it green.

Then we're going to come to the image tab in Higgs field here. and we're going to select GPT image 2. Make sure that we're on 4K.

And here we can paste in the prompt that Gemini gave us. So as you can see, it's going to give us an image like this. Now I can drag that image into Seedance 2 .5.

So I'm just going to write, turn image 1 into a motion graphic. I want the two bars to slide up, and I want the mouse to hover over the green button and click it. And at the end, I just write, keep the text perfectly intact.

If this was more complex, I would get Gemini to write me a new prompt for this specifically, but because it's pretty simple, I'm just writing this. So here it gave us something like this. It took about three minutes.

And if I were to do this in an editing software, it would probably take me like 20 minutes. This one was just a simple one to start, and now we can get a little bit more advanced. So same thing.

I'm going to save this image as, upload it to Gemini, deconstruct this image and write me a prompt for it, make sure all the words are in English, and change a few elements so that it's custom to me. Now once we paste this here into Higgs field, we're going to get an image like this. So now I'm going to drag that into here.

And again, I'm just going to write, turn image 1 into a motion graphic. I want the pictures to pop in with a bouncy animation. And then the black string attaches and connects them all.

Keep all the text high quality and intact, no blurriness. And I'm going to set the duration to 6 seconds. And then hit generate.

And here's the results it gave us. As you can see, none of the text is messed up or blurry or anything like that. Moving on to kind of a clean widget animation, we're going to save this as an image.

We're going to upload it into Gemini and same thing, ask it to generate as a prompt and change the images inside the widgets. I'm also adding here that I want text above the first image and text below the second image, saying house one and house two with an arrow. And then once we paste that into GPT image two, it's going to give us an image like this.

Now we can upload that image into Seedance 2 .5. And here I just wrote again, turn image one into a motion graphic. I want the widgets to pop in with a bouncy animation and I want the text to write out letter by letter with a bouncy text animation.

And again, I always add key. all of the words high quality and intact going to hit generate so here's what it gave us as you can see there's almost no way to tell that this is ai so now for this motion graphic we can get a little bit cinematic here i'm going to save this as an image and i'm going to upload this into gpt image 2.

with this one i don't need a crazy prompt i'm just going to write using this style of chess piece i want you to create me a cinematic environment i want a gray background with cinematic lighting and i want words that say never found beside it with an arrow pointing and it's going to give us something that looks like this so now i'm going to grab that and upload it into cdance 2 .5 turn image 1 into a motion graphic i want the horse piece to rotate up and i want the letters to animate in with a bouncy text animation keep the words high quality and intact no blurriness and i want smooth camera motion this one i'm going to keep it four seconds and hit generate and then here is what it gave us so now we can try one without using a direct reference and we can just gather a bunch of tiny references and give it a crazy prompt and see what it gives us.

So I'm going to come to Pinterest here.

And here I can begin screenshotting just a bunch of references. So we'll go with this one as well. Screenshot that.

And then we can screenshot this one. And then we'll also screenshot this one. Now here in Gemini, I'm going to describe what kind of animation I want and just give it as much detail as possible.

Now we're going to paste that into Higgs field. along with all of our references. And we're going to make sure to add our references.

So for example, at image one here, at image two, at image three. And keep in mind, you can do up to 50 references now. So once we have everything ready, we can select our time.

For this one, I'm going to do a bit longer maybe and hit generate. As you can see, this is more than good enough to use in your own projects or your client's projects. So give it a try yourself and let me know in the comments if you think it's good enough for you to use in your own workflow.

And then also let me know what you guys want to learn next.

The Hook

The bait, then the rug-pull.

The video opens on a flat claim: every image thrown at Seedance 2.5 has animated into a clean motion graphic with zero mistakes, turning five hours of manual keyframing into five minutes of prompting.

Frameworks

Named ideas worth stealing.

00:59concept

Two Paths to a Motion Graphic

Reference-image workflow: pick or generate a still image, then describe exactly how it should move
Prompt-first workflow: ask a chat AI for a fully detailed generation prompt with no reference image at all

Frames every example in the video as a choice between animating something you already have a clear picture of, versus letting a chat model invent the visual from a description.

Steal fordeciding how to brief any AI motion-graphics generation

01:39model

The Three-Tool Prompt Chain

Chat AI (Gemini, Grok, or Claude) deconstructs a reference image and writes a detailed image-generation prompt
Image model (GPT Image 2 or Nano Banana Pro) renders the still at 4K from that prompt
Seedance 2.5 animates the still from a short motion instruction that explicitly says to keep the text intact

The repeated production pattern behind every example in the video, run five separate times with different reference material.

Steal forany still-to-motion AI content pipeline

CTA Breakdown

How they asked for the click.

VERBAL ASK

00:30link

“Seedance 2.5 just dropped a couple of hours ago in Higgs field... so that's what we're gonna be using, the link for that is down below.”

A single soft mention early in the video pointing to the affiliate link in the description (higgsfield.ai/s/zanehoyer-oYRBJc), reinforced only by an end-of-video ask to try the workflow and comment.

MENTIONED ON CAMERA

00:24toolHiggs Field (Cinema Studio) ↗

FROM THE DESCRIPTION

PRIMARY CTAWhere the creator wants you to go next.