Files

3.1 KiB

AI4ArtsEd ComfyUI Nodes - Dev Log

2026-01-19

Photo Prompt Randomizer - Film Types & Editable Prompt

Film Type Selector:

  • Dropdown with 14 analog film stocks + random option
  • Slide films: Kodachrome, Ektachrome
  • Color negative: Portra 400/800, Ektar 100, Fuji Pro 400H/Superia, CineStill 800T
  • B&W: Ilford HP5/Delta 400/FP4/Pan F/XP2, Kodak Tri-X 400
  • {film_description} placeholder in prompt gets replaced by selected film

Editable System Prompt:

  • System prompt now visible and editable in node UI
  • Users can customize prompt generation behavior
  • {film_description} placeholder for film type injection

Improved Default Prompt:

You are an inventive creative. Your task is to invent a REALISTIC photographic image prompt.

Think globally. Avoid cultural clichés. Avoid "retro" style descriptions.
Describe contemporary everyday motives: scenes, objects, animals, nature, tech, culture, people, homes, family, work, holiday, urban, rural, trivia, details.

Choose either unlikely, untypical or typical photographical sujets for realistic photographic images. Be verbose, provide intricate details.

Always begin your output with: "{film_description} of".
Transform the prompt strictly following the context if provided.

NO META-COMMENTS, TITLES, Remarks, dialogue WHATSOEVER, STRICTLY FOLLOW THE INSTRUCTION.

UI Cleanup:

  • Removed debug field (unnecessary clutter)
  • Final inputs: random_seed, film_type, system_prompt, model, unload_model, api_key (opt), context (opt)

2026-01-18

Photo Prompt Randomizer Node

Problem: ComfyUI executes nodes only when inputs change. Integer nodes can be randomized but don't connect to string inputs.

Solution: New node ai4artsed_photo_prompt_randomizer:

  • INT input random_seed for cache-breaking (connected to Integer Randomizer)
  • IS_CHANGED() returning float("nan") forces re-execution
  • LLM generates photographic prompts with Kodachrome style

System Prompt:

You are an inventive creative. Your task is to invent a REALISTIC photographic
image prompt. Choose either unlikely, untypical or typical photographical sujets
for realistic photographic images. Be verbose, provide intricate details.
Always begin your output with: "An Kodachrome film slide of".
Transform the prompt strictly following the context.
NO META-COMMENTS, TITLES, Remarks, dialogue WHATSOEVER, STRICTLY FOLLOW THE INSTRUCTION.

Workflow:

[Integer Randomizer] --INT--> [Photo Prompt Randomizer] --STRING--> [T5-CLIP]

Model List Update

Added new models to both prompt_interception and photo_prompt_randomizer:

Model Price Tag
anthropic/claude-sonnet-4.5 $3.00/$15.00 multilingual
openai/gpt-oss-120b $0.50/$2.00 reasoning
openai/gpt-oss-20b $0.10/$0.40 reasoning

All existing models retained for backwards compatibility.

Git SSH Fix

Switched remote from HTTPS to SSH for persistent authentication:

git remote set-url origin git@github.com:joeriben/ai4artsed_comfyui_nodes.git

Earlier Changes

See git log for history:

git log --oneline