v6: bake start_comfy + claude_start.sh into /workspace

Add claude_start.sh launcher script

Add start_comfy launcher script

Upload folder using huggingface_hub

Upload workflows/api/qwen-edit-restore-v1.json with huggingface_hub

Upload workflows/qwen-edit-restore-v1.json with huggingface_hub

Upload models/loras/qwen_edit_2511_restore_plxrstyle_v1.safetensors with huggingface_hub

Add qwen-edit-turbo-v4 test outputs (mode 1/2/3)

Add qwen-edit-turbo-v4 (API format)

Add qwen-edit-turbo-v4: 3-way sampling mode (4/8-step Lightning, quality) + one-click ref image 2 toggle

Move Qwen_LoRA_Skin_Fix_v2 to models/loras/ (v3 loras workflow)

Move Qwen_LoRA_Amateur_Photo_v1 to models/loras/ (v3 loras workflow)

Add SCAIL-2 test output (extended, 157 frames)

Add SCAIL-2 test output (base 81-frame chunk)

convert_workflow.py: resolve unwired promoted subgraph inputs from sibling widget defaults

Add SCAIL-2 character replacement workflow (API format)

Add SCAIL-2 character replacement workflow (UI format)

Add SCAIL-2 model: sam3.1_multiplex_fp16.safetensors

Add SCAIL-2 model: wan2.1_SCAIL_2_DPO_lora_bf16.safetensors

Add SCAIL-2 model: Wan2_1_VAE_bf16.safetensors

Add SCAIL-2 model: wan2.1_14B_SCAIL_2_fp16.safetensors

Add API version of qwen-edit-turbo-v3-loras

Add qwen-edit-turbo-v3-loras: v3 with subgraphs + user LoRA support

Add Qwen_LoRA_Amateur_Photo_v1.safetensors

Add Qwen_LoRA_Skin_Fix_v2.safetensors

Upload qwen-edit-turbo-v3.json

Delete workflows/qwen-edit-turbo-v3.json

qwen-edit-turbo-v3 rev2: upscale/downscale_input_to_max_size + upscale_output_to_selected_size toggles on both image chains (tested)

Add qwen-edit-turbo-v3 API format (tested: lightning toggle, adaptive/manual output size, output scale, 2-image ref)

Add qwen-edit-turbo-v3: 2511 + Lightning toggle + input preprocessing (QwenEditUtils) + output size control

Add qwen-edit-upscale-2k workflow (UI+API) and Qwen-Image-Edit-2511-Upscale2K lora

docker v5: boot refresh mirrors workflows/ from the repo (refresh-workspace.sh), claude alias (IS_SANDBOX=1 --dangerously-skip-permissions); document interim v4 (no auto-start, script refresh)

download_missing_models.sh: always sync workflows/ from the repo before planning (add --no-refresh to opt out) — stale baked-in workflows made the plan miss newly added models

Add Krea 2 two-image i2i workflows (turbo + raw)

Add klein_9b_enhancer_v2 and klein_snofs_v1_4 LoRAs

Add Krea 2 models: raw bf16, turbo bf16, qwen3vl text encoder, darkbrush LoRA

Add Krea 2 workflows (raw/turbo t2i + turbo i2i, optional LoRA placeholder)

Add full-precision qwen_2.5_vl_7b text encoder (for qwen-edit-turbo-v2)

Add Qwen-Image-Edit-2511-Lightning 8-step LoRA (for qwen-edit-turbo-v2)

Add qwen-edit-turbo-v2.json (full-precision text encoder + 8-step Lightning LoRA)

Add qwen-edit-turbo-v2.json (full-precision text encoder + 8-step Lightning LoRA)

Add remaining API-format workflows; converter fixes (subgraph promoted inputs, rgthree loras, undeclared seed-control widgets)

Add API scripts (upload/download/run/convert) and API-format headswap workflow

Upload 3 files

Delete workflows/qwen-loras-very-good-nsfw.json

Delete workflows/qwen-edit-nsfw.json

Delete workflows/qwen-loras-very-good-nsfw-v2.json

Delete workflows/qwen-loras-very-good-nsfw-v3.json

v3: exact-mirror qwen-edit-nsfw settings (same negative prompt), Qwen4Play at 0.65 — settings verified identical by structural diff

v3: restore second reference-image input (dropped upstream in qwen-edit-nsfw), expand negative prompt