L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Productivité · SKILL.md

workflow_neta_yume_v3_5_txt2img_fullhd_landscape

workflow_neta_yume_v3_5_txt2img_fullhd_landscape

Workflow wrapper imported from `Neta Yume v3.5 txt2img - FullHD Landscape.json`.

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentation de la Skill

GitHub

name: workflow_neta_yume_v3_5_txt2img_fullhd_landscape description: > Workflow wrapper imported from Neta Yume v3.5 txt2img - FullHD Landscape.json. homepage: https://github.com/steliosot/comfy-agent metadata.clawdbot.os: ["darwin", "linux"] metadata.clawdbot.requires.bins: ["python3"] metadata.clawdbot.requires.env: ["COMFY_URL", "COMFY_AUTH_HEADER"] metadata.clawdbot.files: ["workflow.json", "skill.py", "skill.yaml", "scripts/*"] metadata.clawdbot.tags: ["comfyui", "workflow", "library", "txt2img"] metadata.clawdbot.category: "media-generation" metadata.clawdbot.input_type: "text" metadata.clawdbot.output_type: "image/png" metadata.clawdbot.output_can_feed_into: [] metadata.clawdbot.accepts_input_from: [] metadata.clawdbot.priority: 60

workflow_neta_yume_v3_5_txt2img_fullhd_landscape

Imported workflow skill generated from Neta Yume v3.5 txt2img - FullHD Landscape.json.

Family

  • txt2img

Inputs

  • Optional runtime overrides supported by run(...):
    • prompt
    • negative_prompt
    • width, height
    • seed, steps, cfg
    • sampler_name, scheduler, denoise
    • server, headers, api_prefix

Outputs

  • Returns JSON with:
    • status
    • prompt_id
    • output_images

Model Requirements

  • checkpoint: NetaYumev35_pretrained_all_in_one.safetensors -> models/checkpoints

Custom Node Requirements

  • None detected.

Links

Routing Metadata

  • Family: txt2img
  • Input modalities: text_prompt
  • Output modalities: image/png
  • Model families: sd3
  • Node count: 12
  • Complexity score: 3
  • Resource profile: medium
  • Estimated runtime: moderate (about 30-120s depending on server)
  • Max latent resolution hint: 1920x1088
  • Max sampler steps hint: 30

Detected Models

  • checkpoint: NetaYumev35_pretrained_all_in_one.safetensors -> models/checkpoints

Detected Custom Nodes

  • None detected.

Runtime Warnings

  • High output resolution detected; generation time/memory use can increase significantly.
Skills enregistrées