This guide explains the models currently available and the jobs each one is best suited to.

Image models​

GPT Image 2​

Best for: polished campaign images, reference-driven work, product integration, and editing.

GPT Image 2 is the default image model in CraftyAI. It handles detailed natural-language direction well and can work with selected products, logos, and reference assets. It is a strong general choice when accuracy and control matter more than speed.

Grok Image Pro​

Best for: direct text-to-image exploration and fresh visual concepts.

Grok Image Pro is useful when you want to begin from a written idea and explore a visual direction quickly. In CraftyAI it is offered as a text-to-image model, so choose GPT Image 2 or Gemini 3 Pro Image when reference-based editing is important.

Gemini 3 Pro Image​

Best for: product references, structured layouts, and image editing.

Gemini 3 Pro Image responds well to explicit instructions about subject, layout, lighting, background, and exclusions. It supports reference and editing workflows, making it a good alternative for product-focused compositions.

Nano Banana 2 Lite​

Best for: fast text-to-image drafts and straightforward concepts.

This lightweight Gemini model is designed for speed. Use it to test a direction or generate a simple visual from a concise brief. It is exposed as text-to-image only in CraftyAI.

MiniMax Image 01​

Best for: concise text-to-image prompts and quick visual ideation.

MiniMax Image 01 works best with a direct prompt that clearly states the subject, composition, style, lighting, and anything to avoid. CraftyAI automatically keeps its generated prompt within the model's supported length.

Video models​

Whiteboard Explainer​

Best for: narrated explainers, educational content, and hand-drawn stories.

This is CraftyAI's purpose-built workflow for converting a brief or narration into a scene-based drawing animation. It supports landscape and portrait output, optional narration and subtitles, and videos from 5 to 60 seconds.

Seedance 2.0 Fast​

Best for: versatile text-to-video, image-to-video, product motion, and Paper Collage.

Seedance is CraftyAI's default video model and the broadest all-round option. It supports both written prompts and starting images, multiple aspect ratios, and clips from 4 to 15 seconds. It is also the verified animation model behind the Paper Collage workflow.

MiniMax Hailuo 2.3​

Best for: expressive subject movement, atmosphere, texture, and smooth camera motion.

Hailuo supports text and image-to-video generation. Choose it for more dynamic shots where movement and mood are central. CraftyAI currently offers 6- and 10-second clips at 768P or 1080P.

Gemini Omni Flash​

Best for: concise text-to-video scenes with one clear action.

Gemini Omni Flash is available as text-to-video in CraftyAI. It suits an 8-second cinematic instruction with a clear subject, natural camera movement, and a simple ending frame. Starting images are not used with this model on the current interface.

HappyHorse 1.1​

Best for: simple landscape text-to-video shots.

HappyHorse works best with one subject, one action, and one camera move. The current CraftyAI integration produces 5-second landscape clips and does not accept a starting image.

Grok Imagine Video 1.5​

Best for: animating an existing image while keeping its subject and composition recognizable.

Grok Imagine Video is image-to-video only, so a starting image is required. Describe the motion, camera behavior, and atmosphere rather than repeating what is already visible in the frame. Output shape follows the starting image.

How to choose​

Start with the creative requirement rather than the model name:

• choose GPT Image 2 for a dependable image workflow with references and editing;
• choose Nano Banana 2 Lite or MiniMax Image 01 for quick text-only image exploration;
• choose Seedance 2.0 Fast for the most flexible general video workflow;
• choose MiniMax Hailuo 2.3 when expressive movement and atmosphere are the priority;
• choose Whiteboard Explainer for structured, narrated educational content;
• choose Grok Imagine Video 1.5 when the job begins with a specific image that must be animated.

CraftyAI adapts its prompt guidance to the selected model and always shows the available duration, resolution, aspect ratio, and price before generation. The model list will continue to evolve as new options are tested and become reliable enough for the studio.

Choose a workflow​

Create product images with references, or open CraftyAI Video Studio to compare the live duration, resolution, and price options for a video model.

Related reading​

Explore the purpose-built Whiteboard Video and Paper Collage workflows.

Instead of producing a sequence of disconnected clips, Paper Collage plans one coherent story, lets you review that plan, and only then begins generating the finished media.

See Paper Collage in motion​

A tactile way to tell a digital story​

Paper Collage combines torn construction paper, photographic cutouts, halftone newsprint, tape, registration marks, screen-print textures, and oversized editorial typography. Pieces slide, stamp, unfold, and pop into place with restrained parallax and purposeful camera movement.

The result feels crafted rather than synthetic: part magazine cover, part motion poster, and part animated scrapbook.

It works especially well for:

• product and feature announcements;
• short explainers and educational stories;
• creator-led social advertising;
• bold campaign messages and memorable calls to action.

How the workflow works​

1. Start with the message​

Choose Paper Collage in Video Studio, then describe the topic, audience, hook, key benefit, proof point, tone, and optional call to action. You can write freely or use the guided brief as a starting point.

Videos can run from 5 to 60 seconds, giving you enough room for a quick social hook or a more complete product explanation.

2. Add an optional reference​

Upload a product photo, person, or other visual reference when identity matters. CraftyAI preserves defining details and treats the subject as a photographic paper cutout or sticker inside the composition instead of redrawing it into something unrecognizable.

3. Review the plan before generation​

CraftyAI first prepares a scene-by-scene collage strategy. No paid media is generated at this point.

You can read the proposed direction, add feedback, and confirm it when the hook, pacing, and final message feel right. This deliberate pause makes it much easier to catch a weak story before committing to generation.

4. Build the final collage​

Once confirmed, CraftyAI generates the source material, assembles the animated scenes, and keeps the visual language consistent across the complete video. If voice is enabled, narration is written to fit the selected duration; captions follow that narration verbatim.

A few tips for a strong result​

Keep the central message singular. Paper Collage is at its best when every scene pushes toward one memorable idea. Short headlines, a clear visual subject, and a specific final action generally work better than several competing claims.

When using a reference image, choose one with a clean, recognizable subject. The clearer the source, the more confidently it can become the hero cutout within the collage.

Paper Collage is available now in CraftyAI Video Studio.

Related reading​

For a hand-drawn alternative, explore Whiteboard Video, or compare the video models available on CraftyAI.

Give it a narration script, a simple brief, or a reference image. CraftyAI breaks the material into scenes, creates the drawings, and composes them into a finished video with optional voice and subtitles.

See Whiteboard Video in action​

Explain one idea at a time​

Whiteboard animation works because it reveals information gradually. The viewer sees an idea take shape instead of being presented with a finished, crowded frame.

That makes the format a natural fit for:

• product education and feature walkthroughs;
• lessons, concepts, and step-by-step explanations;
• short social stories;
• diagrams, comparisons, and visual summaries.

Three ways to begin​

General Whiteboard​

Enter a narration script or describe the idea you want to explain. CraftyAI turns the material into a concise sequence of drawings designed around the selected duration.

Story Whiteboard​

Use a scene-based brief when sequence and storytelling matter. Each scene can include its own narration and drawing direction, and CraftyAI follows those scenes in order without forcing a predefined narrative arc.

Reference Whiteboard​

Upload a reference image when the subject must remain recognizable. The uploaded image becomes the visual source of truth: CraftyAI preserves important proportions, colors, markings, materials, and identity while translating the subject into a clean whiteboard drawing.

For photographs, Adapt to Whiteboard converts the source into the drawn style. For an illustration that is already suitable, Animate As-Is keeps its existing appearance.

Choose the right output​

Whiteboard Video supports:

• 5 to 60 second videos;
• landscape 16:9 and vertical 9:16 formats;
• 480P, 720P, and 1080P output;
• Standard or Detailed drawing quality;
• optional subtitles;
• Voice Off, Man, Woman, Old Man, or Old Woman narration.

Scene count scales with duration so the animation stays readable rather than rushing through too many drawings. A longer video has room for more scenes; a short one stays focused.

What happens after you press Generate​

CraftyAI prepares the whiteboard runtime, generates the required scene artwork, creates narration when enabled, and composes everything to the exact requested duration and dimensions.

Before the video is delivered, the workflow checks that:

• the output dimensions match the selected resolution and aspect ratio;
• audio and video remain in sync;
• the final duration matches the request;
• subtitles are present only when enabled and stay readable on a single line;
• narration does not overrun the available time.

These checks turn Whiteboard Video from a collection of generated assets into a dependable finished workflow.

Writing a better whiteboard brief​

Use short spoken sentences and pair each one with a concrete drawing instruction. “Explain our delivery process” is workable; “Draw a parcel moving through three labeled checkpoints while the narration explains each step” gives the system a much clearer visual plan.

For reference-led videos, say which details make the subject recognizable. For story videos, keep each scene responsible for one beat or idea.

Whiteboard Video is available now in CraftyAI Video Studio.

Related reading​

Explore the tactile Paper Collage workflow, or compare the video models available on CraftyAI.

More than a changelog​

A release list tells you what shipped. We want these notes to preserve the useful part: the trade-offs, failed approaches, performance discoveries, and tiny interface details that made a feature feel right.

Our goal is simple: make the work legible while it is still fresh.

Expect practical notes from image generation, video workflows, prompt tooling, and the infrastructure holding it together. We will not publish private customer data, credentials, or claims that cannot be supported by the shipped product.

What we intend to publish​

Product decisions​

Some posts explain why a control behaves a particular way or why a workflow has a deliberate constraint. Prompt workflows that remember context, for example, describes why a generated prompt belongs to a specific group of settings rather than remaining valid after every change.

Creative workflows​

Practical guides focus on completing a real marketing task. They should help a reader plan inputs, choose settings, avoid common mistakes, and judge an output—even if that reader arrived through a nonbranded search. Our Whiteboard and Paper Collage articles introduce the formats; step-by-step guides connect those formats to repeatable work.

Model behavior and cost​

Model names are less useful than knowing what each option accepts and what it is suited to. We will document supported text-to-image, image-to-video, duration, resolution, and reference workflows without pretending every model is interchangeable. The CraftyAI model guide is the living starting point, while predictable video pricing explains how the interface and backend calculate cost.

Performance and infrastructure​

Creative tools handle large files and long-running jobs, so small implementation decisions can directly affect trust. We will share useful lessons about media delivery, validation, failure handling, and keeping browser state aligned with backend rules. These posts may be technical, but they should still explain the user consequence.

Release notes versus practical guides​

A release note starts with a shipped change: what became available, how it works now, and any important limits. Introducing Whiteboard Video and Introducing Paper Collage are examples.

A practical guide starts with a reader’s goal. It should remain useful beyond a single release and provide a process that can be repeated. Tags and descriptive titles will make the distinction easier to see, while related-reading links connect an announcement to the deeper workflow behind it.

Engineering notes sit between the two. They explain a design or architecture choice with enough implementation detail to be credible, then translate it into a consequence a user can understand.

Why return​

Prospective users can see how the product actually behaves before investing time in a workflow. Marketers can pick up repeatable ways to prepare references, prompts, and output settings. Developers can follow the trade-offs behind billing, caching, media performance, and generation orchestration.

Publishing close to the work also creates a useful discipline for us. If a feature is difficult to explain clearly, its interface or contract may need another pass. If an article cannot verify a claim against source or tested behavior, that claim should not be published.

What comes next​

We will publish concise build notes alongside deeper technical write-ups and task-focused guides. Each post will be tagged, searchable, dated, and updated when the underlying behavior changes. The aim is a useful record, not a stream of announcements for their own sake.

Explore CraftyAI​

Create product-ready marketing images, or begin with the articles below to understand the models and workflow decisions behind the studio.

Start exploring​

Read about prompt workflows that remember context or browse the models available on CraftyAI.

The choices that affect video cost​

The model is the first dimension. Different video models support different durations, resolutions, aspect ratios, and per-second rates. Duration then determines how many seconds are billed. Where a model offers multiple resolutions, the selected resolution determines the applicable per-second rate.

This means there is no trustworthy universal “price per video.” A short lower-resolution clip and a longer high-resolution clip can use different calculations even when the written prompt is identical. CraftyAI exposes only the options supported by the selected model and updates the displayed estimate as those options change.

One backend source of truth​

Video Studio requests the current model list from /api/video-models. That response contains the selectable durations, resolution options, aspect ratios, defaults, and per-second prices the interface needs. Keeping this data on the backend avoids maintaining a second authoritative price table in browser code.

The server still recalculates every charge before generation begins. It resolves the requested model and validates the submitted duration, resolution, and aspect ratio against its own configuration. A stale or modified browser therefore cannot become the billing authority.

charge = price_per_second_microdollars * duration_seconds

The user’s balance is checked before the provider request, while the successful generation path performs the debit. Failed or rejected generation should not consume the video charge. This separation matters because an interface estimate is helpful, but only server-side validation can safely protect the account balance.

How starting images change the workflow​

A starting image changes prompt and aspect-ratio behavior, but it does not bypass pricing. The backend detects whether an image is present and applies the chosen model’s supported image-to-video rules.

When a starting image is uploaded, CraftyAI uses that image’s shape and ignores the separate aspect-ratio value. The interface disables the aspect-ratio selector and shows Image size to make that constraint visible. Duration and resolution remain relevant where the model supports choices for them, so they still influence the estimate and server calculation.

Not every model accepts a starting image. The available options and interface state depend on the selected model; for example, some models are text-to-video only, while Grok Imagine Video requires an image. The current model guide describes these workflow differences.

Why resolution does not invalidate the prompt​

Resolution affects pixel dimensions, output quality, and potentially price. It does not change the subject, action, camera movement, or story described by a video prompt. For that reason, changing only resolution keeps the current built prompt visible.

Model, duration, aspect ratio, and starting-image state do affect the creative instruction, so those settings participate in the video prompt cache key. If you return to a previous combination, CraftyAI can restore the prompt created for it. This reduces repeated work without mixing incompatible context.

A practical way to control cost​

Start by choosing the model for the required job, not by selecting the highest output setting. Use the shortest duration that can communicate one complete idea. Test motion and composition at the lower available resolution when that output is suitable, then move up only when the destination needs it.

Before generating, check four things beside the button: model, seconds, resolution, and displayed total. If a starting image is present, confirm its shape matches the intended placement. Keep the same creative prompt while comparing resolutions, but rebuild after changing the model or timing.

This approach makes early iterations less expensive and reserves higher-resolution generations for directions that have already proved useful.

Estimate your next video​

Compare CraftyAI image and video models, then open Video Studio to see the live options and calculated total for your chosen setup.

Related reading​

Compare the video models available on CraftyAI, then see the workflow in Whiteboard Video.

Why prompt context becomes stale​

A useful creative prompt encodes more than the sentence in the text box. It reflects the chosen model, the intended placement, selected references, and other controls that affect how the request should be phrased. If a marketer builds a prompt for a square product image and then replaces the product reference or switches models, reusing the old prompt can silently carry the wrong assumptions into the next generation.

The safer mental model is to treat a generated prompt like a settings-specific artifact. It belongs to the combination that produced it. CraftyAI stores that relationship so the interface can distinguish a current prompt from one built for an earlier setup.

What changes an image prompt​

In Image Studio, the creative description, social platform, selected style, model, and selected reference assets all affect prompt construction. Changing one of those inputs makes the visible prompt stale. The main action returns to its build state so the user can generate with current context or rebuild the text first.

Changing the model matters because each image model has its own prompt guidance. Changing a product or logo reference matters because the prompt may describe how that asset should appear. A platform change can alter composition and placement guidance. These are creative changes, not merely output settings.

The image dimensions also influence placement guidance when CraftyAI generates an image, but a user can still write and submit a manual prompt directly. Building a prompt is an aid, not a gate that must be passed before generation.

What changes a video prompt​

Video prompts are cached against the selected model, duration, aspect ratio, and whether a starting image is present. Those choices change the story the prompt needs to tell. A six-second image-to-video shot requires different instruction from a longer text-to-video sequence.

A resolution change is deliberately different. Resolution affects output size and price, but it does not change the subject, motion, camera direction, or narrative. CraftyAI therefore keeps the current video prompt when only resolution changes. This avoids asking users to rebuild identical creative direction merely because they selected a different output quality.

When a starting image is added, the image defines the output shape and CraftyAI disables the separate aspect-ratio choice. Removing or replacing that starting point changes the creative context and should trigger a fresh prompt decision.

Restoring useful work without hiding changes​

When the current settings match a previously built combination, CraftyAI can restore that prompt. Switching from setup A to setup B hides A’s prompt; switching back to A brings it back. Clearing a prompt removes that cached entry. This makes experimentation less repetitive while keeping it obvious which text belongs to which controls.

The visible button state reinforces the distinction. After a prompt is built, the action becomes Rebuild Prompt. A prompt-affecting change restores the normal build state unless that exact combination already has a cached result.

A practical workflow for marketers​

1. Choose the model, placement, style, and reference assets before refining the text.
2. Build the prompt, read it, and correct any product or brand details.
3. Generate a small test rather than changing several variables at once.
4. If you change a reference, model, duration, or aspect ratio, review the prompt again.
5. Treat resolution as an output and cost choice unless the composition itself must change.

The goal is not to make prompts precious. It is to make the relationship between a brief and its context visible, so a fast iteration does not accidentally become a stale one.

Put the workflow into practice​

Create marketing images from a product photo and keep each prompt tied to the references and settings that produced it.

That distinction keeps the workflow responsive without pretending that old context is still current.

Related reading​

See how the same backend-first approach makes video pricing predictable, or review the models available on CraftyAI.