ByteDance's Seedream and Seedance models, inside Syntina.
Change only the region of an image you select, or turn one render into video with sound up to 30 seconds. We use ByteDance's Seedream 5 Pro and Seedance 2.5 in Syntina's Layer editor, on the canvas and in Multishot.
- Image models
- Seedream 5 Pro and Seedream 5.5
- Video models
- Seedance 2.5 (4-30 s) and Seedance 2.0 Pro (4-15 s)
- Where
- Canvas, Layer editor, Multishot and the mobile app
- Credits
- Seedream 5 Pro 9 (Standard) or 18 (2K); 5.5 4 or 7, +2 per image; Seedance 2.5 4 s 720p 303
Latest from ByteDance
What ByteDance shipped in recent months
Compiled from the ByteDance Seed team's own blog posts. ByteDance's own evaluations are reported as ByteDance's claims.
13 February 2026
Seedream 5.0 Lite
Focused on reasoning. It transfers style and colour from a reference image, edits from vague instructions and preserves unedited areas better. Optional real-time web search can be switched off for stability. ByteDance calls the model relatively small and says structure, realism and aesthetics still have room to improve.
Source: ByteDance Seed blog: Seedream 5.0 Lite8 July 2026
Seedream 5.0 Pro
Region editing: mark an area with a point, lasso, box or doodle and change colours, swap materials, add or remove objects. It splits posters into more than 10 independent layers with transparency. It handles information visualisation (infographics, charts, timelines) and text in over ten commonly used languages.
Source: ByteDance Seed blog: Seedream 5.0 Pro31 July 2026
Seedance 2.5
Up to 30 seconds of video in one generation (2.0 was 15 seconds), extendable to several minutes with extension rounds. It generates audio and video together. It takes up to 30 images, 10 video clips and 10 audio clips in one pass, including "clay render" references: textureless 3D models that set composition and camera.
Source: ByteDance Seed blog: Seedance 2.59 November 2026
Seedream 5.5
ByteDance says unedited regions stay pixel-identical across 100 successive edits and that it outputs transparent (RGBA) images natively. Its own text-corruption figures: 0.4% severe, 1.7% minor. A Pro version was announced as coming. We added it to Syntina on 17 November 2026.
Source: ByteDance Seed blog: Seedream 5.5
Why ByteDance
Where these models stand out in an architect's work
Instead of putting Seedream and Seedance side by side in a list, we looked at the four abilities an architect actually needs.
Region editing
In Seedream 5.0 Pro you mark the area with a point, lasso, box or doodle; the model changes the colour or material, adds an object or removes one.
For an architect that means swapping a facade cladding or a floor material, or taking an object out of the scene.
Information visualisation
According to ByteDance, Seedream 5.0 Pro also produces images that carry information, such as infographics, charts and timelines, and writes text in over ten commonly used languages.
That can help with labels, scale and diagrams on a presentation board.
Untouched areas that stay put across edits
According to ByteDance, Seedream 5.5 keeps untouched regions pixel-identical across 100 successive edits. In a scene with many revision rounds, that is what limits drift.
In Syntina 5.5 is the fast editing model: it returns in about 25 seconds and carries a "Fast" tag on the canvas.
Long video with sound and references
Seedance 2.5 can produce up to 30 seconds in one generation and creates sound together with the picture. It offers timestamp-level control over narrative, camera and pacing, and per-clip edits after generation.
Given a textureless 3D model as a reference, it can use it for composition and camera. That makes a raw 3D export interesting for architects.
For the clip above we gave a room render as the first frame and asked for a slow camera move toward the chair by the window: Seedance 2.5, 720p, 5 seconds, no sound.
How we use it in Syntina
We did not add the models to a list. We put them at the right step of the work.
For each model we looked at where it helps most and offered it there. The reasons are below.
ModelArk first, with a fallback route
We reach the models through BytePlus's ModelArk service. If the first route refuses or fails, we have a fallback route so the job is not lost; generation continues without you noticing.
Region editing in the Layer editor
We put Seedream 5 Pro in the Layer editor as region editing. You mark an area with a box; the model sees the whole image and changes only that region. It takes about 2 minutes and costs 9 credits. There is no reference image input.
The still engine in Multishot
In Multishot we use Seedream 5 Pro (2K) for still frames. We want a consistent set of frames from the same references. Seedream 5 Pro accepts up to 10 image inputs.
Seedance 2.5 is flagged new on the canvas
We put Seedance 2.5 near the top of the video models with a "new" flag. It offers 4-30 seconds, start and end frames, native audio on by default and 480p, 720p or 1080p; the source image sets the aspect ratio. A 4-second 720p video costs 303 credits.
Seedance 2.0 Pro for narrative presets
Narrative and cinematic presets default to Seedance 2.0 Pro; you can switch to 2.5. Reference-to-video in Multishot also runs on 2.0 Pro. A 4-second 720p video costs 194 credits.
In the mobile app too
The mobile app has image editing with Seedream 5 Pro and video with Seedance 2.5 and 2.0.
We cut the Seedream line to two
On 17 November 2026 we added Seedream 5.5 and left only 5 Pro and 5.5 in the picker. Work saved with Seedream 4.5 or 5 Lite moves to 5.5 automatically. 5.5 costs 4 credits at Standard and 7 at High, plus 2 per input image, and is also in the SketchUp and 3ds Max plugins. 5 Pro is 9 at Standard (~1.5K) and 18 at 2K. We also retired Seedance 2.0 Mini; Seedance 1.5 Pro is no longer recommended.
For architects
Where it helps in the office
Mark the facade, swap the material
Select a facade section of the render with a box and write "timber cladding instead of the light grey render". You show the client an alternative material without regenerating the whole image.
A walkthrough video with sound from a render
Use a finished render as the start frame and turn it into a walkthrough with sound up to 30 seconds. It can be a short social clip or a transition for a client presentation.
A consistent shot set from the same references
If you want a set of frames of one space from different angles, give your references to Multishot once so the frames look alike.
Presentation boards with labels
For a board image with charts, scale bars or section labels you can use Seedream 5 Pro's information visualisation. Always check the text by eye before delivery.
Limits
What it does not do, and what to watch for
Region editing regenerates the whole image
The model changes the area you mark but regenerates the whole image, and it takes about 2 minutes. Small differences can appear outside the selection; compare the result with the original.
Complex motion is a known weak spot
In its own post ByteDance lists the physical plausibility of complex motion and scene stability with several interacting subjects as areas still to improve. For crowded or very busy scenes, expect to try a few times.
Seedance 2.5 tops out at 1080p in Syntina
The highest resolution for Seedance 2.5 on the canvas is 1080p. If you need more, upscale the result separately.
Images with real people
Images that contain real people can be caught by the provider's privacy filter. If a scene needs a human figure, describe the person instead of using a photo of a specific individual.
FAQ
Frequently asked questions
- Do I need a separate account to use ByteDance's models in Syntina?
- No. The models run inside Syntina; generations are deducted from your Syntina credits.
- What is the difference between Seedream 5 Pro and Seedream 5.5?
- According to ByteDance, 5 Pro focuses on region editing, layered output and information visualisation, while 5.5 focuses on keeping untouched areas intact across successive edits. In Syntina 5 Pro costs 9 or 18 credits; 5.5 is the fast model at 4 or 7 credits plus 2 per input image.
- Is Seedream 5.5 in Syntina?
- Yes. ByteDance announced it on 9 November 2026 and we added it to the canvas and the SketchUp and 3ds Max plugins on 17 November.
- Should I pick Seedance 2.5 or 2.0?
- 2.5 makes 4-30 seconds and 2.0 Pro 4-15 seconds. A 4-second 720p video costs 303 credits on 2.5 and 194 on 2.0. 2.0 Pro suits short, narrative clips; 2.5 suits longer shots with sound.
- Does the video have sound?
- In Seedance 2.5, native audio is on by default and is generated together with the picture.
- Can I give a reference image for region editing?
- Not for now. Seedream 5 Pro region editing in the Layer editor has no reference input; you describe the change in text.
Sources
Where the company facts on this page come from
- ByteDance Seed blog: Seedream 5.0 Lite2026-02-13
- ByteDance Seed blog: Seedream 5.0 Pro2026-07-08
- BytePlus ModelArk: Seedream 5.0 Pro tutorial
- ByteDance Seed blog: Seedream 5.52026-11-09
- ByteDance Seed blog: Seedance 2.52026-07-31
- BytePlus blog: Seedance 2.52026-08-06
- ByteDance Seed: Seedance 2.0
Sources checked on 14 November 2026.
ByteDance and its product names are trademarks of their owners. Syntina uses these models through ByteDance's public API; this page was not prepared by ByteDance and does not imply endorsement or a formal partnership. The company's own measurements and benchmarks are reported as its claims.