[PR #9] [MERGED] Enable krea2_turbo_t2i, z_image_turbo_t2i and qwen_image_2_1_edit #9

New issue

Closed

opened 2026-10-31 22:06:12 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-31 22:06:12 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/Comfy-Org/comfyci-runpod-worker/pull/9
Author: @marawan206
Created: 10/18/2026
Status: ✅ Merged
Merged: 10/18/2026
Merged by: @marawan206

Base: main ← Head: enable-image-model-workflows

📝 Commits (1)

710bf85 Enable krea2_turbo_t2i, z_image_turbo_t2i and qwen_image_2_1_edit

📊 Changes

1 file changed (+3 additions, -3 deletions)

View changed files

📝 manifest/workflows.json (+3 -3)

📄 Description

What

Flips enabled to true for the three image-model workflows staged in #5:

Workflow	Golden	Noise floor	Exec (4090)	VRAM peak
`krea2_turbo_t2i`	v0.42.0	identical (r1 == r2)	61 s	18.1 GB
`z_image_turbo_t2i`	v0.42.0	identical (r1 == r2)	68 s	20.6 GB
`qwen_image_2_1_edit`	v0.42.0	identical (r1 == r2)	80 s	14.4 GB

Their models have been on the network volume since 2026-10-09 (sync run 37950773949), so this PR needs no sync and no new secrets.

Goldens

Candidates were first generated at v0.39.2 on 2026-10-09 and left unblessed. Since then 90 commits landed on master (v0.40.0 to v0.42.0), including the Qwen 3.x text-encoder performance work in #17007 and two comfy-kitchen bumps, so the goldens were regenerated at v0.42.0 (the tag master sits on today) before blessing: golden run https://github.com/Comfy-Org/comfyci-runpod-worker/actions/runs/39171392479, bless run https://github.com/Comfy-Org/comfyci-runpod-worker/actions/runs/39172188006.

All three v0.42.0 outputs are byte-identical to the v0.39.2 candidates (same sha256 and size), so no drift to accept: these models have been stable across that range on the cu128 lane. current.json for each now points at v0.42.0 with the reason recorded.

Effect

Every new master commit now runs five workflows instead of two. Per run that adds roughly 4 to 5 minutes of RTX 4090 time; at the recent rate of 4 to 9 tested commits a day the regression endpoint goes from well under $1/day to about $1 to $2/day.

The four templates in #8 are independent of this change and merge cleanly either side of it (checked with git merge-tree).

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/Comfy-Org/comfyci-runpod-worker/pull/9 **Author:** [@marawan206](https://github.com/marawan206) **Created:** 10/18/2026 **Status:** ✅ Merged **Merged:** 10/18/2026 **Merged by:** [@marawan206](https://github.com/marawan206) **Base:** `main` ← **Head:** `enable-image-model-workflows` --- ### 📝 Commits (1) - [`710bf85`](https://github.com/Comfy-Org/comfyci-runpod-worker/commit/710bf85aa7b5c4c51d1c45cab2aea97d4f9dda48) Enable krea2_turbo_t2i, z_image_turbo_t2i and qwen_image_2_1_edit ### 📊 Changes **1 file changed** (+3 additions, -3 deletions) <details> <summary>View changed files</summary> 📝 `manifest/workflows.json` (+3 -3) </details> ### 📄 Description ## What Flips `enabled` to `true` for the three image-model workflows staged in #5: | Workflow | Golden | Noise floor | Exec (4090) | VRAM peak | |---|---|---|---|---| | `krea2_turbo_t2i` | v0.42.0 | identical (r1 == r2) | 61 s | 18.1 GB | | `z_image_turbo_t2i` | v0.42.0 | identical (r1 == r2) | 68 s | 20.6 GB | | `qwen_image_2_1_edit` | v0.42.0 | identical (r1 == r2) | 80 s | 14.4 GB | Their models have been on the network volume since 2026-10-09 (sync run 37950773949), so this PR needs no sync and no new secrets. ## Goldens Candidates were first generated at v0.39.2 on 2026-10-09 and left unblessed. Since then 90 commits landed on master (v0.40.0 to v0.42.0), including the Qwen 3.x text-encoder performance work in #17007 and two comfy-kitchen bumps, so the goldens were regenerated at v0.42.0 (the tag master sits on today) before blessing: golden run https://github.com/Comfy-Org/comfyci-runpod-worker/actions/runs/39171392479, bless run https://github.com/Comfy-Org/comfyci-runpod-worker/actions/runs/39172188006. All three v0.42.0 outputs are byte-identical to the v0.39.2 candidates (same sha256 and size), so no drift to accept: these models have been stable across that range on the cu128 lane. `current.json` for each now points at v0.42.0 with the reason recorded. ## Effect Every new master commit now runs five workflows instead of two. Per run that adds roughly 4 to 5 minutes of RTX 4090 time; at the recent rate of 4 to 9 tested commits a day the regression endpoint goes from well under $1/day to about $1 to $2/day. The four templates in #8 are independent of this change and merge cleanly either side of it (checked with `git merge-tree`). --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-31 22:06:12 +00:00