Skip to content

Releases: suyunxing/dsh-draw-mode

v2.0.1

Choose a tag to compare

@suyunxing suyunxing released this 25 Oct 02:27

中文

v2.0.1 — README 与发布说明改为中英双语

仅文档改动,代码与 2.0.0 完全一致。

  • README 现在是两份完整文档:中文在前、英文在后,两边内容等价(功能、环境要求、安装、加工作流、自动识别规则、sidecar、模型文件名映射、指向你的 ComfyUI、面板、20 项配置表、yaml 示例、备注)。
  • v2.0.0 的发布说明也已补上中文。

English

v2.0.1 — bilingual README and release notes

Documentation only; the code is identical to 2.0.0.

  • The README is now two complete halves: Chinese first, English second, covering the same ground (what it does, requirements, install, adding workflows, slot detection, the sidecar file, model file mapping, pointing it at your ComfyUI, the panel, the 20-option table, the YAML example, notes).
  • The v2.0.0 release notes were given a Chinese half as well.

Full Changelog: v2.0.0...v2.0.1

v2.0.0

Choose a tag to compare

@suyunxing suyunxing released this 25 Oct 00:48

中文

v2.0.0 — 空框架

破坏性变更: 自带的工作流库已移除。

  • 不再有工作流库,不再有模型族清单。 原先自带的十个 ComfyUI 工作流、manifest.json,以及 Anima / Krea 2 / Qwen-Image 2.1 那张对照表全部删除。把你自己的 Save (API Format) JSON 放进 workflowsDir,agent 就能跑它。
  • 自动识别输入位置。 逐个分析工作流,找出提示词、负向词、宽、高、步数、cfg、seed、denoise、输入图和 LoRA 分别在哪。只有字面量输入才会被当作可替换位置,连到别的节点的值不会被覆盖。可以放 <工作流名>.draw.json sidecar 覆盖识别结果(label、task、slots、defaults)。
  • 通用 LoRA 注入。 图里已有 LoraLoader 就复用它,否则在模型产出节点和采样器之间插入一个,并把 clip 连线改接过去。
  • 所有路径都可配置。 comfyRoot / comfyPython / comfyMain / comfyWorkdir / comfyArgs / modelsRoot / workflowsDir / outputDir,可在侧栏面板里改并保存到 ~/.dsh/draw-mode/settings.json。文本框留空则回落到插件配置;清除覆盖 会删掉该文件。
  • draw_models 改为读 ComfyUI 通过 /object_info 实际报告的模型文件,不再用硬编码清单;draw_reverse 不再需要指定目标模型族。
  • 空闲关闭仍是可选(idleShutdownMs: 0),且默认作用域为 spawned,只会停掉它自己拉起的 ComfyUI。

从 1.x 升级:自带的工作流 id 已不存在。请用 Save (API Format) 导出你自己的图。


English

v2.0.0 — empty framework

Breaking: the bundled workflow library is gone.

  • No workflow library, no family list. The ten bundled ComfyUI graphs, manifest.json, and the Anima / Krea 2 / Qwen-Image 2.1 table are removed. Drop your own Save (API Format) JSON into workflowsDir and the agent can run it.
  • Automatic slot detection. Each graph is inspected to find where prompt, negative, width, height, steps, cfg, seed, denoise, input images and LoRA live. Only literal inputs become slots, so nothing wired to another node is overwritten. A <name>.draw.json sidecar can override the detection (label, task, slots, defaults).
  • Generic LoRA injection. Reuses an existing LoraLoader when the graph has one, otherwise splices one between the model producer and the sampler and rewires the clip link.
  • Every path is configuration. comfyRoot / comfyPython / comfyMain / comfyWorkdir / comfyArgs / modelsRoot / workflowsDir / outputDir, editable from the sidebar panel and saved to ~/.dsh/draw-mode/settings.json. Empty fields fall back to the plugin config; clear overrides deletes the file.
  • draw_models reads the model files ComfyUI actually reports via /object_info instead of a hard-coded list; draw_reverse no longer takes a target family.
  • Idle shutdown stays opt-in (idleShutdownMs: 0) and is scoped to spawned by default, so it only ever stops a ComfyUI it started itself.

Upgrading from 1.x: the bundled workflow ids no longer exist. Export your own graphs with Save (API Format).

Full Changelog: v1.0.1...v2.0.0

v1.0.1

Choose a tag to compare

@suyunxing suyunxing released this 23 Oct 08:54