ByteDance Seedream 4: lock this still before you buy motion (4K, 3cr)
ByteDance Seedream 4 is the stills row. Animate after approval. Do not generate the pack inside the video model.
ByteDance Seedream 4 is the stills row. Animate after approval. Do not generate the pack inside the video model. The catalog calls it the previous-generation ByteDance image model: still a capable, lower-cost text-to-image pick when you do not need 4.5's stronger prompt comprehension. Content type is image. Credits are 3. Max output is 4K. No duration. No audio. Requires no image. You are buying a still you can reject for 3 credits, not a clip you cannot un-generate.
Seedream 4.5 exists and is published. Use it when the prompt is dense. Use 4 when the brief is a pack of 4K frames you will approve by eye, then hand to an image-to-video row. The ByteDance roster has both. This page is only 4.
Previous generation is a price, not a verdict
"Previous-generation" in the description is a routing hint. It is not an instruction to skip the model. 3 credits at 4K is how you fill a review board: hero, alternate crop, colorway, thumbnail. Styles on the row include Realistic, Artistic, and Anime. Aspects include 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3, and 21:9. Qualities include 1K, 2K, and 4K. That is a stills toolkit, not a video toolkit.
The text-to-image tool is where you run it. The best image generator ranking is the rest of the category. Do not open a 70-credit video model to "see if the product looks right." That is how teams burn a day's credits on unapproved packaging.
Approve the still. Then buy motion.
A video model will invent a bottle if you let it. It will also invent a slightly different bottle on take two. The cheap fix is to lock the SKU here, at 3 credits, 4K, and only then animate. The expensive fix is to prompt the video row until the label is almost right.
Once the frame is the one you would print, send it to image-to-video. Keep the still in the kit. The next ad set should start from the same file, not from a new text prompt that almost describes it.
If the still needs a surgical change — swap the cap, kill a reflection — that is an edit model, not another Seedream 4 roll and not a video prompt. Do not use motion as a retoucher.
4K is the ceiling of this row, not a clip length
There are no supports_durations. Asking Seedream 4 for "a 5-second shot" wastes the prompt. You will get a still. If you needed 5 seconds, you needed a video category after approval.
requires_image is false. You can start from text. max_reference is 10 if you do feed references, but the category is still text-to-image. Do not confuse a style hint with a video first-frame.
Three credits. 4K. Previous generation on purpose. Lock the pack. Then spend the video budget on a frame the client already signed.
FAQ
Why not start in the video model and screenshot the best frame?
Because the video model is not a 3-credit stills machine. Seedream 4 is. Approve composition, type, and product at 4K for 3 credits, then animate the winner.
When should I use Seedream 4.5 instead?
When the prompt itself is the hard part — nested instructions, tight layout, copy the description calls 4.5 stronger at. For volume 4K boards where you will pick by eye, 4 at 3 credits is the listed job.
Does Seedream 4 make audio or clips?
No. Content type is image. Audio is empty. Durations are empty. It is a still.
What resolution should I export for later animation?
The row tops out at 4K, with 1K and 2K as quality options. Export the ratio you will animate. A 9:16 still is a 9:16 first frame later. Do not lock a 1:1 hero and then fight the crop in video.