Seedream 5.0 Flash takes 14 references; Sume seedream-5-lite takes 10
Seedream 5.0 Flash lists up to 14 input references on OpenRouter. Sume's catalog caps references at 10 and lists seedream-5-lite, not Flash.

Seedream 5.0 Flash accepts up to 14 input_references on its OpenRouter page, which is four more than Sume's cap. Sume's Image API allows at most 10 references on a catalog model (16 on ChatGPT Image 2.5), and its catalog carries bytedance-seed/seedream-5-lite, not Flash. If your brand board has 11 to 14 images, Flash is the one with room. On Sume you cut the set to 10.
Side by side
The Flash column is from the OpenRouter listing and the Higgsfield changelog, both read on 2026-10-05. The Sume column is from the Image API docs.
| Item | Seedream 5.0 Flash (OpenRouter) | Sume bytedance-seed/seedream-5-lite |
|---|---|---|
| Max references | 14 (input_references) | 10 |
| Resolution | 1K or 2K | Read the catalog resolution descriptor |
| Aspect ratios | "1:1" to "21:9" | Per-model catalog list |
| Seed | Supported | Not served in v1; returns 400 unsupported_parameter |
| Launch note | Higgsfield changelog, Sep 23: fast tier, multi-reference style blending | Same family, different row; Flash is not in the Sume catalog |
What the reference cap means in practice
A reference cap is a hard limit. On Sume, a request with an input_references array longer than the catalog's descriptor is rejected, and a model whose descriptor is {"min": 0, "max": 0} takes no references at all. Check the descriptor first, because it changes per model.
- Rank your references. Keep the ten that carry the style: palette, texture, one hero subject.
- Merge small swatches into one board image before you upload. One board image uses one slot.
- Reference URLs must be public HTTPS; Sume rejects localhost, private-network and non-HTTPS URLs before submission.
Check the live descriptor
Read the catalog to see the number Sume will enforce for the row today, rather than trusting a blog post (including this one).
import asyncio, os, httpx
async def main():
headers = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
async with httpx.AsyncClient(timeout=30) as c:
r = await c.get("https://api.sume.com/v1/images/models", headers=headers)
r.raise_for_status()
for m in r.json()["data"]:
if m["id"].startswith("bytedance-seed/"):
print(m["id"], m["supported_parameters"].get("input_references"))
asyncio.run(main())Pricing note
The OpenRouter text we fetched shows no price for Flash, so this post states none. Sume's static rate table lists seedream-5-lite at $0.035 before margin; at provider list x 1.25 that is $0.04375 before rounding. Quote the live pricing line from the endpoints call for the billed figure. For the resolution angle on Flash, see the earlier Seedream 5.0 Flash comparison.
Decision rule
Count your references. At 10 or fewer, the cap is irrelevant and you can choose on price, speed and output. At 11 to 14, Flash fits the set as is, while Sume needs you to merge or drop images. Keep in mind that more references is not automatically better: each extra image is one more thing the model has to reconcile, so a tight set of five often steers more cleanly than fourteen. We have not measured this on Flash, so treat it as a test to run, not a finding.
Sources
Related posts
More in Models
- Sound-on Reels product spot in one call: Omni Flash, 8 s, $1.00
Reels default to sound on. Omni Flash 1.1 returns synced audio: 8 s at 720p is $1.00 on Sume, against $3.27 for a silent clip plus voice and music.
- Spain vs Mexico, Brazil vs Portugal: regional voiceover on MAI, Sume
MAI-Voice-2.1 lists es-ES, es-MX, pt-BR and pt-PT. Sume's guard compares only the primary language. How to pick a regional read and test it for 1 cent.
- Text edits on Sume: use sume/auto or pin ideogram/ideogram-v4.5?
Sume's sume/auto never says which model ran, so a text-edit workflow that depends on Ideogram 4.5 behavior should pin ideogram/ideogram-v4.5.
- Sume image models that do not list the 4:5 Instagram ratio
Imagen 4, Grok Image and Higgsfield Soul do not list 4:5. Which Sume ids do, what 4:5 means in pixels, and the nearest ratio to use when a model lacks it.
Written by Sume