Make someone stop scrolling. Start with a clear hook, a person or an action worth watching.
Explore the modelsOur recommended starting order, with alternatives for different needs.
Selection reviewed
Tell the idea in one sequence
Build a campaign scene up to 30 seconds around your visual references. Plan the hook, action and closing image before generating.
Output goes up to 1080p. References guide the result; they do not guarantee an identical face or object.
Give references a clear role
A 1080p option for image-led scenes, reference-guided work and video transformation. Plan what each reference should control.
Inputs and audio behavior differ between modes. A storyboard is direction, not a promise that every cut will match.
Create for a larger screen
Generate 5–15 second clips with audio, with output options up to 4K and support for image, video and audio references.
Audio is generated with the clip. 4K describes output size, not a guarantee of better motion.
Direct the beginning and end
Use opening and ending frames, mix references or extend a clip, with generation lengths up to 30 seconds.
For video input and extension, source plus generated duration must fit within the 30-second limit.
Revise a short scene
Create 3–10 second scenes, guide them with images, or edit and extend a supplied clip, with output options up to 4K.
Availability is limited. MCP editing needs a supplied source; conversation-based retakes stay in the website workspace.
Start with Seedance 2.5 for a longer campaign scene, or Kling Omni Pro for a storyboard-led spot. H3 covers short scenes with generated sound, Wan 3 Prime adds source-based variants, and Omni Flash 1.2 suits short revisions. Add exact logos and claims during editing.
This is an editorial order for starting your tests, not a measured quality ranking. We consider the available inputs, duration, audio and output options for this task. The clips use different prompts and settings; a single result cannot establish which model is best. Check the full model details before you create.
Pick a model, add your prompt or reference, then check the quote.
Bring a rough idea. Develop the direction in conversation and approve the quote for each creation.
Explore the MCP connections for your assistant, with setup and capabilities for each host.
Studio assistance and media generation have separate costs. Always review the current generation quote. Pricing →
Main Character turns a fictional yellow backpack into an 18-second ad. The assistant generated the source shots through MaxVideoAI and handled the edit, titles and mix. You can direct this kind of production in the conversation without opening an editing application; the documented examples use Codex and its available tools. The fictional product is not a test of exact merchant-SKU fidelity.
Watch Main Character →Make someone stop scrolling. Start with a clear hook, a person or an action worth watching.
Open the approved starting image → watch the exact original result.
This MaxVideoAI-owned demonstration uses a generated, unbranded watch image approved for public use. The brief was a restrained product reveal: keep the cobalt strap, metallic case and lime dial details readable, with a fixed camera and a cobalt/lime light sweep behind the watch. Wan 3 Prime suited this setup because it accepted the same image as both first- and last-frame references for the intended return to the starting pose; the motion still needed human review.
| Recorded setting | Accepted render |
|---|---|
| Model | Wan 3 Prime |
| Input | Same approved image as the first and last frame |
| Format and duration | 16:9 requested · 6 seconds |
| Resolution | 1080p requested · original file: 1918 × 1080 |
| Audio | Off; the published clip has no audio track |
| Motion direction | Fixed orientation and camera; slight vertical lift and gentle return; stationary pedestal |
Quote and approval. In the September 2026 demonstration, the owner approved the exact $2.08 USD media quote before confirmation and accepted the completed Wan result after playback. An earlier completed FLUX.3 attempt at $2.16 USD was rejected for its movement quality: the two approved attempts cost $4.24 USD in media credits. These are historical amounts, not today's quotes or an assistance bill. No new generation was made for this guide.
What to inspect. The blue strap and metallic case remain recognisable while the coloured light moves behind the product. The motion is restrained, so this is a hero reveal rather than a demonstration of the watch's functions. Inspect the small dial markings at full size before using it in an ad. The image has no brand logo: this case does not establish exact logo preservation. Owner acceptance is a creative review of this clip, not a general benchmark win for the model.
Continue in Studio. Choose Open Studio, sign in, and start the product-ad conversation. Add your own authorised product image and describe the format, motion and details you need to preserve. Ask for Wan 3 Prime with the recorded settings, review the currently supported reference controls and a fresh exact quote, then confirm that quote before generating. Sol assistance credits are separate from the media quote. Download the individual finished clip from your library; this example does not promise an assembled timeline-film export.
Start with Seedance 2.5 for a longer campaign scene, or Kling Omni Pro for a storyboard-led spot. H3 covers short scenes with generated sound, Wan 3 Prime adds source-based variants, and Omni Flash 1.2 suits short revisions. Add exact logos and claims during editing.
Use Seedance 2.5 or Wan 3 Prime for complete product stories, MiniMax H3 for high-resolution delivery, and Kling O3 Pro for structured reveals.
Usually yes. Keep the model and reference pack stable while changing one hook, offer, or ending at a time.
Compare the models
Useful side-by-side pages for validating tradeoffs before picking a model.
Explore another use case
