Skip to content

v1.2.0

Latest

Choose a tag to compare

@rookiestar28 rookiestar28 released this 26 Oct 07:20
· 1 commit to main since this release

RookieUI 1.2.0 adds native MOSS-Image generation and experimental Layout guidance, improves sidebar startup, and includes the security fixes and tooling updates released since v1.1.2.

MOSS-Image Generation

  • Added MOSS-Image Base and MOSS-Image SFT text-to-image presets using native ComfyUI encoding, sampling, and FLUX.2 latents.
  • Added model-role validation and host prerequisite checks for the diffusion model, Qwen3.5 9B generation encoder, VAE, required nodes, and sampler. Missing or incompatible selections fail before submission.
  • Added Skip Layer Guidance (SLG) controls: enabled for Base, disabled for SFT, and enabled for both Layout modes unless explicitly overridden.
  • Saved PNG metadata preserves MOSS generation parameters, model selections, SLG, and the original Layout data for PNG Info restoration to Txt2Img.

Experimental Layout Guidance

  • Author up to 32 ordered object, character, or text boxes, with a preview that follows the selected output dimensions.
  • Edit normalized box coordinates, literal rendered text, captions, and optional color palettes. Invalid layouts block generation with actionable feedback.
  • Layout drafts survive disabling the control, profile changes, and supported parameter restoration. Empty or disabled layouts use the ordinary authored prompt.
  • Layout remains experimental: exact placement and text rendering are not guaranteed.

Sidebar Reliability and Compatibility

  • Sidebar startup and client identity handling tolerate blocked or unavailable browser storage, including the host-sidebar and legacy-launcher paths.
  • Added source-contract checks for ComfyUI Core 0.42, Frontend 1.59.1, and workflow templates 0.11.96, while retaining the separate general and Qwen Image 2.1 compatibility scopes.
  • Real GPU checks covered MOSS Base/SFT generation and Layout, preview actions, and saved PNG restoration using int8_convrot diffusion models, qwen3.5_9b_int8_convrot.safetensors, and flux2-vae.safetensors. The tested frontend was an official build from Frontend 1.59.1 source CI.
  • Existing Stable Diffusion generation and native Flux.2 Klein PNG workflow replay were also checked on that host. Legacy A1111-form restoration for Klein remains unsupported.
  • Updated development tooling to Vite 8.3.5.

Security Fixes Included Since v1.1.2

  • Prompt Workbench translation and AI assist use only the official OpenAI and MyMemory HTTPS endpoints. Custom endpoints, redirects, unsafe network destinations, and environment proxy routing are blocked; TLS certificate verification remains enabled.
  • Existing custom endpoint configurations require an explicit switch to the official endpoint before execution. API keys stay masked in exports and browser responses.
  • The optional embedding-directory input accepts only ComfyUI-registered embeddings folders or their subfolders. Other filesystem paths are rejected instead of scanned.
  • Translation accepts language codes and reads local dictionaries only from its catalog folder. Translation errors no longer expose file paths or contents.
  • XYZ Plot grid assembly uses exact generated file names within the ComfyUI output folder.

Validation

  • The Windows Full Gate passed secret detection, all pre-commit hooks, dependency auditing, architecture checks, frontend type validation, 1,116 backend tests with 18 conditional skips, 286 frontend unit tests, and 43 Playwright E2E tests.
  • GPU validation is scoped to the MOSS model combinations and host described above; the full test gate does not imply live coverage of every model variant.

Defaults and Upgrade Notes

  • MOSS sampling defaults are 1024 x 1024, 25 steps, CFG 2.5, er_sde / simple, denoise 1.0.
  • MOSS does not reproduce every official template default: prompt enhancement is unavailable, negative prompts start empty, seeds start at -1 for random generation, and automatic encoder selection prefers int8_convrot, then BF16, then w4a8_gptq. The official templates preselect w4a8_gptq and enable prompt enhancement for ordinary text-to-image.
  • Automatic VAE selection prefers flux2-vae_small.safetensors, then flux2-vae.safetensors. BF16 and w4a8_gptq model selections were not live-qualified.
  • MOSS requires the corresponding host nodes and model files. It currently excludes img2img, LoRAs, Hires, ControlNet, ADetailer, and A1111 prompt compilation. See the MOSS preset notes and the official model and Apache 2.0 license.
  • Update RookieUI through ComfyUI Manager or run git pull --ff-only in the extension directory and reinstall requirements.txt with the ComfyUI Python environment. Restart ComfyUI and reload the browser afterward. Registry availability depends on its review status; use the manual update path if Manager still offers an older release.
  • Third-party or local OpenAI-compatible provider endpoints remain unsupported. Check the official endpoint and credentials before using translation or AI assist after upgrading.

Full changelog: v1.1.2...v1.2.0