Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

Runway image backend: add Seedream 5 (pro/lite) models with ordered reference images

Closed
  • Is something wrong?
#13,199 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

@atomantic is already working on this.

Since Oct 23, 2026.

  • #13218 by @atomantic — merged

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
48/100
Issue type
Feature
Clarity
Clearly specified
Activity status
Active
Tech stack
javascript, node.js
Domain
ai, backend

Research direction

Start with lib/runwayImageModels.js, then trace the visual-conditioning compiler and render flows for Image Gen, locked FableLoom scenes, and universe reference-sheet and place renders. Check the @runwayml/sdk text-to-image types and Runway pricing guide for supported ratios and credits. Done when both models support text-only renders, ordered reference mapping, selection in the named flows, and the requested unit tests and provenance fields.

Written by the indexing model from the issue text.

Description

enhancement

Follow-up to #13105 (merged in #13117). The new Runway image backend (lib/runwayImageModels.js) ships gen4_image and gen4_image_turbo only. The addendum on #13105 asked for Runway's Seedream models too, but it arrived after that work started.

Ask

Add Runway's ByteDance Seedream text-to-image models to the curated catalog. Make seedream5_pro selectable as the default for character identity-pack, reference-sheet, place and storyboard renders.

model ratios (per @runwayml/sdk text-to-image types) references
seedream5_pro 1024:1024, 1184:896, 896:1184, 1376:768, 768:1376, 1296:864, 864:1296, 2048:2048, 2304:1728, 1728:2304, 2720:1530, 1530:2720, 2496:1664, 1664:2496, auto_1k, auto_2k optional referenceImages[]
seedream5_lite 2048:2048, 2304:1728, 1728:2304, 2848:1600, 1600:2848, 2496:1664, 1664:2496, 3136:1344, 3072:3072, 3456:2592, 2592:3456, 4096:2304, 2304:4096, 3744:2496, 2496:3744, 4704:2016 optional referenceImages[]

Why: in side-by-side tests on stylized, non-human characters with a specific costume, Seedream held identity and costume much better than the local defaults (which drift to a generic real-world animal and drop the costume). It's the model I'd pick for identity packs.

Notes

  • Reference syntax differs. Seedream references aren't @tag-addressed like Gen-4's. The catalog row needs a per-model reference mode (tagged vs ordered), and the visual-conditioning compiler has to map compiled bindings accordingly (ordered references plus a prompt preamble naming what each one is).
  • Capabilities. Declare reference support and the max count per model, so locked FableLoom renders can admit Seedream.
  • Credits. Verify per-image credits against Runway's pricing guide and record them in the row, like the Gen-4 rows.
  • Aspect mapping. Map a FableLoom 16:9 render to the nearest Seedream ratio (e.g. 1376:768 or 2720:1530) and record the effective ratio in provenance.

Acceptance criteria

  • Image Gen → runway lists seedream5_pro and seedream5_lite, and renders text-only stills with each.
  • A locked FableLoom scene with an approved identity pack renders through seedream5_pro, sends the pack anchor as a reference, and records the model and reference mode in provenance.
  • Universe reference-sheet and place renders can pick runway + seedream5_pro.
  • Unit tests cover the request body per Seedream model and the ordered-reference mapping (no live calls).
Dominant language
JavaScript
Stars
38
Forks
32
Avg merge
41m
Merged PRs (30d)
991

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from atomantic/PortOS

All issues in atomantic/PortOS

Similar issues

More JavaScript issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.