Feat[models]: Add support for Qwen Image#9003

Open

lstein wants to merge 27 commits intoinvoke-ai:mainfrom

lstein:feat/qwen-image-2512

Commits on Mar 28, 2026

refactor: rename QwenImageEdit -> QwenImage base type

lstein
and
claude
committed
feat: add QwenImage variant type to distinguish edit vs txt2img models

lstein
and
claude
committed
feat: add Qwen Image 2512 txt2img support

lstein
and
claude
committed
chore: ruff & lint:prettier
lstein
committed
fix: remove unused frontend exports (zQwenImageVariantType, isQwenImageEditMainModelConfig)

lstein
and
claude
committed
fix: make QwenImage variant optional to fix model detection tags

lstein
and
claude
committed
fix: restore Qwen Image Edit starter models with distinct variable names

lstein
and
claude
committed
fix: restore correct GGUF filenames in Qwen Image Edit starter model URLs

lstein
and
claude
committed
fix: skip reference images in graph for non-edit Qwen Image models

lstein
and
claude
committed
fix: only set zero_cond_t=True for edit-variant GGUF models

lstein
and
claude
committed
fix: recall Qwen Image advanced params (component source, quantization, shift)

lstein
and
claude
committed
fix: remove unnecessary async from QwenImageComponentSource parse

lstein
and
claude
committed
fix: prevent Flux LoRAs from being detected as Qwen Image LoRAs

lstein
and
claude
committed
chore: ruff
lstein
committed
fix: don't force reference image to output aspect ratio in VAE encoding

lstein
and
claude
committed

Commits on Apr 1, 2026

fix: generate noise in float32 instead of float16

lstein
and
claude
committed