Seedream 5.0 Lite 生成結果
字節跳動 Seedream 5.0 的快速高性價比版本:約 10 秒生成或編輯圖像,最高支援 2K 解析度,最多支援 10 張參考圖,固定每張僅需 3 點數。
Seedream 5.0 Flash is the speed tier of ByteDance Seed's Seedream 5.0 image family. It reached developers on BytePlus ModelArk in September 2026 as model version dola-seedream-5-0-flash-260915, about ten weeks after Seedream 5.0 Pro. ByteDance's documentation sums up the split in one line each: Pro "provides higher image quality", while Flash "generates images faster and costs less, making it suitable for latency- and cost-sensitive scenarios."
Flash is not a cut-down feature set. ByteDance documents the same capabilities for both tiers: text-to-image, editing from one or several reference images (up to 10), region edits guided by marks or coordinates, layer decomposition, and native text in 14 additional languages, from Arabic and Thai to Vietnamese and German. Both output at 1K, 1.5K or 2K. What Flash trades away is the extra rendering effort that Pro spends on fine detail.
On Flux1 AI, Flash costs 每張 3 點數 at any size, with no charge for reference images, and most results are back in about 10 seconds. That makes it the model for the part of a project where you need many attempts: testing compositions, trying ten backgrounds for one product, or checking that an idea works before you spend more on Seedream 5.0 Pro or GPT Image 2.5.
What Flash is built for, and how the speed and the flat price change the way you work.
Most runs come back in 10–20 seconds at every size, 2K included. Seedream 5.0 Pro takes 1.5–3 minutes for the same request. At Flash's pace you can change one phrase, run it again and compare the two results side by side, which is how most images actually get good.
3 credits buys a 1K, 1.5K or 2K image, and references are free whether you upload one or 10. There is no reason to draft small and re-render later: generate at 2K from the first attempt and keep whichever result works.
Upload a product photo, a face, a palette or a layout and say what to keep and what to change: swap a background, change an outfit, place a product in a new scene, or combine subjects from several photos. Refer to images by number ("the bag from image 2") so the model knows which reference does what. ByteDance also documents edits guided by marks drawn on the image, such as a circle around the area to change.
ByteDance lists native text generation in 14 additional languages for the Seedream 5.0 Pro and Flash pair, including Japanese, Korean, Arabic, Thai, Vietnamese, Spanish, Portuguese, French and German. Keep the copy short and quote it exactly; for dense, multi-tier layouts and infographics, ByteDance positions Pro as the quality tier.
Modern e-commerce banner: Today's Hot Sale. Product centered, price $19.99 below, clean minimal style.
Based on this image, create 1 image. Maintain the same style and elements, but the layout of the elements can be adjusted.
Four steps from idea to download. Credits are only spent when a run succeeds.
In the panel above, write the subject, setting, lighting, framing and any exact text in quotes. One clear sentence per idea works better than a long list of style tags.
Upload up to 10 images: a product, a person, a style or a layout. The price stays at 3 credits however many you add.
2K is selected by default and costs the same as 1K, so leave it unless you need a smaller file. Choose from eight aspect ratios, from 9:16 stories to 21:9 banners.
Results arrive in about 10 seconds and are saved to your history. Change one thing per run, reuse a good result as a reference, and move the winner to Seedream 5.0 Pro if it needs a final quality pass.
Flash is quick enough to test prompts instead of guessing. These rewrites fix the habits that waste the most runs.
All five run on Flux1 AI with the same credits. Flash has the lowest price per image of the five; the others win on output size, speed or polish.
| 模型 | 适用場景 | Max output | References | 典型耗時 | Credits / image |
|---|---|---|---|---|---|
| Seedream 5.0 Flash | Fast drafts, variations and quick edits at 2K | 2K | Up to 10, no extra cost | 10–20 s | 3 |
| Seedream 5.0 Pro | Final artwork, infographics, dense multilingual text | 2K | Up to 10, extras billed | 1.5–3 min | 6–10 |
| Seedream 5.0 (Lite) | Everyday generation and editing, 4K output | 4K | Up to 10 | 5–10 s | 6 |
| Nano Banana 2 Lite | Quick, low-cost edits and previews at 1K | 1K | Up to 10 | 5–10 s | 6 |
| Nano Banana | Everyday conversational edits at 1K | 1K | Up to 10 | 10–15 s | 6 |
Same family, the same feature list in ByteDance's documentation, and the same 2K ceiling. The difference is effort: Pro takes 1.5–3 minutes and costs 6 credits at 1.5K or 10 at 2K, plus a charge for each extra reference, while Flash returns in about 10 seconds for 3 credits flat. Use Flash to find the composition, then rerun the winning prompt and references on Pro when fine detail, small text or dense layouts matter.
Both are the budget, high-speed tiers of their families. Nano Banana 2 Lite is slightly quicker here (5–10 s) but stops at 1K and costs 6 credits. Flash costs 3 credits and goes to 2K, so it is the better pick when you need a full-size image rather than a preview. If you later need Google's model at 4K, Nano Banana 2 starts at 10 credits.
One price for every run. You see the total before you press Generate, and failed runs are refunded automatically.
| Run | What you get | Credits |
|---|---|---|
| Text-to-image | One image at 1K, 1.5K or 2K | 3 |
| Image edit | One image at any size, with 1 to 10 reference images | 3 |
| Failed or blocked run | Credits go back to your balance automatically | Refunded |
| For comparison | Seedream 5.0 Pro, one 2K image with one reference | 10 |
Basic · 300 credits/mo
≈ 100 Flash images at any size
Pro · 800 credits/mo
≈ 266 Flash images at any size
Max · 2000 credits/mo
≈ 666 Flash images at any size
Credits are shared across every image and video model, so Flash drafts, Pro finals and a few videos all come out of one balance.
What this Seedream 5.0 Flash workflow does and does not do, so you can pick the right tool the first time.
Flash stops at 2K here, as it does on ByteDance's own API. For 3K or 4K, use Seedream 5.0 Lite or Nano Banana 2.
ByteDance's API can split an image into a base plus up to 16 transparent layers and accepts coordinate tags for region edits. This generator offers text-to-image and reference editing only, and returns one flattened image per run. To mark a region, draw on the image in any editor before you upload it.
ByteDance positions Pro, not Flash, as the quality tier. Very small text, crowded layouts and fine textures are where you are most likely to see the gap; rerun those images on Seedream 5.0 Pro.
Each run returns a single image and costs 3 credits. To compare variations, run the prompt again or change one detail per run.
Flash draws what you write; it does not look anything up. Quote the exact copy you need, then proofread spelling, prices and brand names before you publish.
Explicit sexual content is not supported. ByteDance's safety filter can block prompts or reference images that break the content policy, and a blocked run is refunded.
It is the fast, low-cost tier of ByteDance Seed's Seedream 5.0 image family. It generates images from text and edits images using up to 10 references, outputs up to 2K, and on Flux1 AI usually finishes in about 10 seconds for 3 credits.
Use Flash for drafts, variations, quick edits and high volume, and Pro for final artwork, infographics and dense multilingual text. ByteDance documents the same capabilities for both, describing Pro as higher quality and Flash as faster and cheaper. Here, Pro takes 1.5–3 minutes and costs 6–10 credits; Flash takes about 10 seconds and costs 3.
Each run costs 3 credits, the same at every size and with any number of references. Credits come with any Flux1 AI plan and work on every model on the site. If a run fails, its credits are returned automatically.
BytePlus added it to its ModelArk platform in September 2026 under the model ID dola-seedream-5-0-flash-260915. It followed Seedream 5.0 Lite (February 2026) and Seedream 5.0 Pro (July 8, 2026).
No. Lite came first, in February 2026, as a reasoning-focused model with optional online search, and on Flux1 AI it outputs up to 4K for 6 credits. Flash belongs to the later Pro and Flash pair: it shares Pro's editing features, stops at 2K and costs 3 credits.
ByteDance offers Flash to developers through BytePlus ModelArk; see the official Seedream 5.0 pro / flash tutorial. The Flux1 AI API currently serves Seedream 5.0 Lite and the Nano Banana models, not Flash. The API docs list what is available.
每次最多 10 張, at no extra cost. Give each image one job (subject, outfit, background or style) and refer to it by number in the prompt; references that compete with each other weaken the result.
No. ByteDance has not released weights for Seedream 5.0 Flash, so there is no official offline version, and "Seedream 5.0 Flash download" files or apps are not from ByteDance. You run it online and download your finished images from your results.
Write a prompt or drop in a reference, and get a 2K image in about 10 seconds.