Skip to content

[feat] image: add MiniMax-H3-Image generation and editing - #280

Merged
princepride merged 1 commit into
ThinkFlowLab:mainfrom
FayeSpica:feat/minimax-h3-image
Nov 10, 2026
Merged

princepride merged 1 commit into
ThinkFlowLab:mainfrom
FayeSpica:feat/minimax-h3-image

Conversation

@FayeSpica

@FayeSpica FayeSpica commented Nov 10, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

  • Add a MiniMax Official Images preset for MiniMax-H3-Image and MiniMax-H3-Image-Highspeed on the MiniMax China endpoint. Canvas and Agent use the same adapter for text-to-image and prompt-based editing with up to 16 ordered PNG, JPEG, or WebP references.
  • Save the provider result URL for download recovery. A timeout during submission leaves the outcome unknown and does not automatically send another billable request.

Linked issue

  • Confirm the linked issue or mark as no issue.

Labels

  • Apply the feature and provider labels.

Test plan

  • Local, Python 3.12.15 and Node.js 26.11.1 on base 6c7f33dc plus the local changes.
    • python -m pytest tests/test_minimax_image.py -q → 3 passed. Covers generation and ordered editing requests for both models, result recovery, and timeout without resubmission.
    • node --test tests/providerPresets.test.ts → 8 passed.
    • npm run build and npm run check:i18n → passed.
  • Left to CI: full regression suite.

Evidence that it works

The six canvas examples below show text-to-image generation and editing with one or two reference images. Each screenshot is followed by the original PNG.

Case Model Operation Output Result
Case 01 · NIGHT GHOST MiniMax-H3-Image Text-to-image 2048 × 2048 PNG adopted on canvas
Case 02 · NEURAL DRIFT MiniMax-H3-Image Text-to-image 2720 × 1536 PNG adopted on canvas
Case 03 · TITAN RISING MiniMax-H3-Image Text-to-image 1536 × 2720 PNG adopted on canvas
Case 04 · INSIDE THE TITAN MiniMax-H3-Image-Highspeed Text-to-image 2048 × 2048 PNG adopted on canvas
Case 05 · RED PROTOCOL MiniMax-H3-Image Edit with Case 01 as reference 2048 × 2048 PNG adopted on canvas
Case 06 · GHOST IN THE TITAN MiniMax-H3-Image Edit with Case 01 and Case 03 as references, in that order 2048 × 2048 PNG adopted on canvas

Case 01 · Night Ghost

  • Case 01 · Night Ghost generated image

Case 01 · Night Ghost

Case 02 · Neural Drift

  • Case 02 · Neural Drift generated image

Case 02 · Neural Drift

Case 03 · Titan Rising

  • Case 03 · Titan Rising generated image

Case 03 · Titan Rising

Case 04 · Inside the Titan

  • Case 04 · Inside the Titan generated image

Case 04 · Inside the Titan

Case 05 · Red Protocol

Case 06 · Ghost in the Titan

  • Case 06 · Ghost in the Titan generated image

Case 06 · Ghost in the Titan

  • Case 05 · Red Protocol generated image
Case 05 _ RED PROTOCOL

Provider verification

  • Live MiniMax calls returned readable PNGs for text-to-image and reference-based editing. Automated tests used mocked HTTP responses to check request routing and download recovery.

Bilingual UI

  • Updated and checked the preset in zh-CN and en-US.

Signed-off-by: Weiming Liao <liaowm5@gmail.com>
@FayeSpica
FayeSpica marked this pull request as ready for review November 10, 2026 16:06
@FayeSpica FayeSpica added type:feat New capability or behaviour area:provider Model providers and generation protocols area:canvas Canvas, nodes and direct manipulation area:agent Agent runtime, assistant chat and tools modality:image Image generation or editing labels Nov 10, 2026
@princepride
princepride merged commit 8a5bf57 into ThinkFlowLab:main Nov 10, 2026
5 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

area:agent Agent runtime, assistant chat and tools area:canvas Canvas, nodes and direct manipulation area:provider Model providers and generation protocols modality:image Image generation or editing type:feat New capability or behaviour

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants