v0.3.41
Comfy-Org/ComfyUIv0.3.41Jun 17, 2025by github-actions[bot]
AI Summary
Introduces Cosmos Predict2 models, fp8 ops for AMD, and a LoRA training node with SQLite database support.
Key Highlights
- Cosmos Predict2 text-to-image and image-to-video models
- fp8 ops for AMD/gfx1201
- LoRA Trainer node
- SQLite database support
New Features
- Cosmos Predict2 models
- fp8 ops
- LoRA Trainer node
- SQLite database support
- Chroma TE support for regular Flux
Full Release Notes
## What's Changed * Let chroma TE work on regular flux. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8429 * chore: add badge and link to social network Twitter by @Olexandr88 in https://github.com/comfyanonymous/ComfyUI/pull/8427 * Alternate fix for #8435 by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8442 * Cleaner torch version comparisons. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8453 * Enable fp8 ops by default on gfx1201 by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8464 * Enable AMD fp8 and pytorch attention on some GPUs. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8474 * Fix WebcamCapture IS_CHANGED signature by @kmewhort in https://github.com/comfyanonymous/ComfyUI/pull/8413 * Apple most likely is never fixing the fp16 attention bug. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8485 * Update template to 0.1.27 by @comfyui-wiki in https://github.com/comfyanonymous/ComfyUI/pull/8487 * Fix black images on some flux models in fp16. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8495 * Fix mistake in last commit. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8496 * auto register web folder from pyproject by @jtydhr88 in https://github.com/comfyanonymous/ComfyUI/pull/8478 * Add support for sqlite database by @pythongosssss in https://github.com/comfyanonymous/ComfyUI/pull/8444 * Revert "auto register web folder from pyproject (#8478)" by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8497 * Allow specifying any frontend semver suffixes by @webfiltered in https://github.com/comfyanonymous/ComfyUI/pull/8498 * Revert "Update template to 0.1.27" by @webfiltered in https://github.com/comfyanonymous/ComfyUI/pull/8502 * Add a warning for old python versions. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8504 * Auto register web folder by @jtydhr88 in https://github.com/comfyanonymous/ComfyUI/pull/8505 * debug -> warning by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8506 * Add '@prerelease' to use latest test frontend by @webfiltered in https://github.com/comfyanonymous/ComfyUI/pull/8501 * Fix Kontext aspect ratio only works as widget by @christian-byrne in https://github.com/comfyanonymous/ComfyUI/pull/8508 * Support cosmos predict2 text to image 2B and 14B models. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8517 * Uncap cosmos predict2 res and fix mem estimation. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8518 * Add cosmos_rflow option to ModelSamplingContinuousEDM node. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8523 * LoRA Trainer: LoRA training node in weight adapter scheme by @KohakuBlueleaf in https://github.com/comfyanonymous/ComfyUI/pull/8446 * Bump embedded docs to 0.2.2 by @comfyui-wiki in https://github.com/comfyanonymous/ComfyUI/pull/8512 * Bump template to 0.1.28 by @comfyui-wiki in https://github.com/comfyanonymous/ComfyUI/pull/8510 * allow extra settings from pyproject.toml by @jtydhr88 in https://github.com/comfyanonymous/ComfyUI/pull/8526 * Small cosmos attention code refactor. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8530 * Generalize SEEDS samplers for RF by @chaObserv in https://github.com/comfyanonymous/ComfyUI/pull/8529 * Support Cosmos predict2 image to video models. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8535 * SaveLora now saves in the same filename format as all the other nodes. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8538 * Add correct eps to ltxv rmsnorm. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8542 * Allow setting min_length with tokenizer_data. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/8547 * Multistep DPM++ SDE samplers for RF by @chaObserv in https://github.com/comfyanonymous/ComfyUI/pull/8541 ## New Contributors * @Olexandr88 made their first contribution in https://github.com/comfyanonymous/ComfyUI/pull/8427 * @kmewhort made their first contribution in https://github.com/comfyanonymous/ComfyUI/pull/8413 **Full Changelog**: https://github.com/comfyanonymous/ComfyUI/compare/v0.3.40...v0.3.41