Seedance 2.5 Is Now Available in Tolstoy

Quick answer: Seedance 2.5 adds longer, reference-controlled video to Tolstoy

Seedance 2.5 is now available in Tolstoy AI Studio for text-to-video, image-to-video, first-and-last-frame, and multi-reference video generation. Its clearest advantage is control over a longer edit: one generation can run from 4 to 30 seconds, use native audio, and work from a much larger reference set than Seedance 2.0.

• Make a complete short-form edit: use up to 30 seconds for a hook, product proof, and closing beat in one generation.
• Bring more references: Tolstoy supports up to 30 image references and 10 audio references with Seedance 2.5.
• Choose the right input mode: start from text, one image, first and last frames, or a reference set.
• Generate sound with the picture: native audio is on by default, with the option to generate a silent clip when you plan to finish sound elsewhere.
• Know the tradeoff: Seedance 2.5 currently outputs 480p or 720p in Tolstoy. Use Seedance 2.0 when the final deliverable must be 1080p.

Availability: Tolstoy added Seedance 2.5 to AI Studio on August 7, 2026.

What is Seedance 2.5?

ByteDance describes Seedance 2.5 as an audio-video joint generation model built for 30-second storytelling, precise reference control, and editing. The official release highlights smoother motion, stronger visual consistency, better interpretation of reference videos, and production controls such as camera movement, performance blocking, white-model control, and green-screen editing.

For ecommerce teams, the gain is continuity. One generation can carry the product, set, visual language, shot order, and audio direction across a longer edit. That makes the model useful when a five-second animation is too short and stitching several unrelated generations would create visible continuity problems.

Seedance 2.5 vs. Seedance 2.0: which should you use?

Choose between Seedance 2.5 and Seedance 2.0 based on the deliverable.

Use Seedance 2.5 when continuity across several beats matters more than maximum resolution. Use Seedance 2.0 for a shorter hero clip that must reach 1080p, or when the brief is simple enough that the larger reference surface adds no value.

Where does Seedance 2.5 help ecommerce teams?

Build a product story with a beginning, proof, and payoff

A 20- or 30-second clip can cover more than a camera move. A skincare concept can open on the problem, move into texture and application, then end on the finished routine. A footwear ad can show the product in motion, cut to material details, and close on the full look without asking three separate generations to match.

Carry more campaign references into one generation

Use separate references for the exact product, palette, location, styling, shot order, and sound. Tell the model what each reference controls. One image can define product appearance, another can set lighting, and a storyboard can set the sequence. Re-describing every image in prose gives the model competing instructions and makes review harder.

Turn a product still into motion-first creative

Start from a clean product or campaign image and specify the motion, camera, environment, and ending state. This is the most direct route for teams that already have approved photography. If you are building that workflow for the first time, use our guide to turn product images into videos with AI as the production checklist around the model.

How do you use Seedance 2.5 in Tolstoy?

1. Open Tolstoy AI Studio and name the model. Tell the Studio agent to use Seedance 2.5 so the model choice is explicit.
2. Choose the input mode. Use text for an open concept, one image to animate an approved frame, keyframes for a defined start and finish, or reference mode when several assets have distinct jobs.
3. Set the deliverable before the scene. Name duration, aspect ratio, placement, and whether the clip needs audio. A 9:16 paid-social cut and a 16:9 PDP hero need different framing.
4. Write the edit in beats. Give each shot a time range, one camera move, the product action, and the transition into the next shot.
5. Add constraints that can be reviewed. Protect product shape, packaging colors, label placement, materials, and any claims that must not change.
6. Generate the shortest useful test first. Review product accuracy and motion before committing to a longer pass.

Copy-and-paste prompt: a cinematic product film

This concept uses one visual tension: gravity behaves normally everywhere except near the product. Water climbs toward it, becomes light, and resolves into a held hero frame. Upload a clean front packshot as @Image1, a three-quarter view of the same product as @Image2, and a dark mineral, water, or lighting reference without other products or readable brand copy as @Image3.

Use Seedance 2.5 in reference mode. If the model or any required reference is unavailable, stop and tell me what is missing before generating.

Creative premise: Gravity breaks around one hero product. A single drop of water travels upward, pulls light out of darkness, and reveals the product as if the room has been waiting for it.

Deliverable: 15-second cinematic product film, 16:9, 720p, 24 fps, with native sound design. No speech, music, captions, or generated on-screen text.

Reference roles: @Image1 is the source of truth for the front silhouette, packaging proportions, colors, cap, and logo position. @Image2 is the source of truth for the three-quarter geometry. @Image3 controls only the black mineral set, wet surface, hard white light, and restrained amber highlight. Do not copy products, text, or logos from @Image3.

0–4 seconds, extreme macro: In near darkness, a single water droplet moves upward across wet black stone against gravity. Follow it with one slow lateral macro slide. The product is present only as a soft, towering silhouette in the background. Use a 100 mm macro-lens feel, very shallow depth of field, realistic surface tension, and one crisp droplet sound.

4–10 seconds, low reveal: Match-cut the droplet highlight into a razor-thin white rim light tracing the real product silhouette from @Image1. Cut to the three-quarter geometry from @Image2 as concentric water ripples travel inward toward the product. Make one controlled 45-degree camera orbit. The set remains almost black; one restrained amber reflection reveals the material finish. Add a low stone resonance and a quiet sub-bass swell.

10–15 seconds, final hero: Hard cut to a centered front packshot. Make one slow push toward the product while the inward ripples stop at its base and the amber reflection narrows into a clean vertical line. Hold the final composition completely still for the last two seconds. Leave intentional negative space on the left for typography added in editing.

Film language: High contrast, deep blacks, controlled specular highlights, restrained halation, subtle 35 mm grain, deliberate focus falloff, physically plausible water, and real lens behavior. The pace is tense and quiet, not glossy, fast, or euphoric.

Constraints: Show exactly one product. Preserve its silhouette, cap, package proportions, colors, logo position, and material finish in every shot. Do not invent or redraw label copy. No hands, people, smoke, floating particles, extra products, subtitles, graphic overlays, neon lighting, warped reflections, morphing geometry, or logo changes.

What should you check before publishing a Seedance 2.5 video?

A longer generation creates more places for a product or subject to drift. Review the whole clip, not only the opening and final frame.

• Product fidelity: silhouette, color, label, closure, texture, and relative scale stay consistent.
• Continuity: product position, lighting direction, props, and motion carry cleanly across cuts.
• Physical errors: hands, reflections, liquid, fabric, and product edges do not warp mid-motion.
• Audio: dialogue, effects, and music match the right beat; replace generated sound in post when exact wording or timing matters.
• Claims and text: add legal copy, prices, product claims, and exact typography after generation so they remain readable and approved.
• Channel fit: the first two seconds work without sound, the crop survives mobile UI, and the final export matches the placement.

Keep readable interface footage, exact logos, and legal copy in your editing layer. Generative video is strongest as the motion and visual layer; it should not be the source of truth for text.

When should you choose another video model?

Choose a different model when the brief has one of these requirements:

• 1080p is mandatory: use Seedance 2.0 standard or another model that supports the target resolution.
• The clip is a simple five-second packshot: a simpler or faster model can do the job without the larger reference surface.
• You need exact UI or typography: use real screen capture and deterministic editing for the truth layer.
• You need a photoreal real-person reference: use a model and workflow explicitly approved for that likeness use case. Seedance 2.5 may reject some real-person references.

Use Seedance 2.5 for longer, controlled stories with several references. Do not use 30 seconds or 30 inputs merely because the fields exist. Add them when they solve a real continuity or art-direction problem.

Create with Seedance 2.5 in Tolstoy

In Tolstoy AI Studio, you can attach the product packshots and set reference, name Seedance 2.5 in the brief, review the generated clip, and keep the approved output with the rest of the campaign media.

Start free with Tolstoy AI Studio and ask the agent to use Seedance 2.5 for your next product-video brief.

Sources and update note: Model positioning and showcase images come from the official ByteDance Seedance 2.5 page. Background on the model family comes from ByteDance’s Seedance 2.0 launch note. Tolstoy availability, input modes, limits, and resolution support were verified against the production model roster on August 7, 2026.