Repository navigation
Orange v1.1
Curated workflows
- Adds LTX 2.5 Video with synchronized audio, 2–12-second lengths, and an optional first-frame image in one tool. Final output defaults are 1152x768 or 1344x768, with portrait support.
- Adds Qwen Image 2.1 Turbo Edit with the native eight-step schedule, madebyollin texture-fix VAE, and an optional second reference image.
- Adds MOSS-Image SFT text-to-image generation with three aspect ratios and text-focused poster examples.
- Combines Klein image modification and multi-image reference editing through an optional second upload.
- Moves prompt enhancement into Orange, with model-specific instructions for LTX, Qwen Edit, and MOSS instead of extra ComfyUI refiner nodes.
- Adds All, Image, Edit, Video, and Tools library filters. Image is reserved for text-to-image workflows; new manifest tags become filters automatically.
- Refreshes Z-Image, Krea, MOSS, and Qwen thumbnails and describes editing/video model strengths in plain language.
Tool configuration
- Reorganizes the editor into Overview, Inputs & workflow, and Check & save, with a generator preview and Tool ID directly below Display Name.
- Keeps forms and workflow connections open. Both optional-input workflows and their image mappings are visible together, with node and field dropdowns populated from each graph.
- Exposes optional-image behavior, alternate workflow uploads/mappings, duration and frame timing, aspect ratios, output scaling, fixed assets, and prompt enhancement for ordinary imported tools as well as curated tools.
- Checks both workflow modes using unsaved editor values, with one Check Workflow button and a separate Save Tool action.
Model locations and backend compatibility
- Resolves model locations per backend on a temporary submission copy, keeping shared workflow files portable.
- Reuses exact paths or unique exact filenames across ComfyUI subfolders and configured model paths, without guessing a different model or precision.
- Adds backend-specific Save model binding controls for duplicate filenames in Preflight and library results. Unique matches resolve automatically; workflow files are never rewritten by binding saves.
- Shares resolution rules across Preflight, library readiness, and generation, including installed tools and optional-input variants.
- Refreshes model inventory during scans and expires cached incompatibility. Explicit queue rejections allow one same-backend retry only when a fresh inventory changes the resolved model paths.
- Supports typed COMBO inventories and native expandable image conditioning slots in Preflight.
Downloads and reliability
- Adds optional Hugging Face download credentials for gated models, with HF_TOKEN support and separate save/remove actions.
- Stores saved tokens in Windows Credential Manager, macOS Keychain, or Linux Secret Service under the Orange server account, with no plaintext fallback. Existing plaintext credentials migrate only after protected storage is verified.
- Keeps authorization on the exact HTTPS huggingface.co origin and strips it from external download redirects.
- Improves MP4 gallery handling and scrolling for longer generation forms.
Upgrading
- Restart through the launcher to install updated dependencies. Existing local tools, workflows, prompts, branding, assets, and model bindings remain deployment-owned; reviewing or reinstalling a curated pack is a separate action.
- Run Check Workflow after upgrading. Different backend folders resolve automatically when model filenames match uniquely. See the workflow guide and portable model locations.
Validation
All 261 Python tests pass on Python 3.10 and 3.12 in GitHub CI, including database tests. JavaScript syntax and Tool Editor round-trip checks pass. Local browser checks cover optional inputs, backend model bindings, one Check Workflow action, and mobile layout.
Qwen and MOSS GPU validation used ComfyUI 0.45.0 on an RTX 5000 Ada. BF16 GPU rendering and fresh model downloads were not exercised. Orange preserves the independently verified managed ComfyUI pin; Preflight checks actual node/model availability on each backend.