STEP 2B — one variation's prompt, out of the set the LLM returned. Appends the identity and scene locks verbatim, so the authored part varies and the fidelity…

Cloud AI in ComfyUI without an API key. Generate gpt-image-2 images and run GPT-5 text + vision through your existing ChatGPT/Codex login, plus Replicate models. 61 nodes: image gen, LLM, video, TTS, and a full product-variation pipeline.

comfyui-arkennemasis

prompts_json

cell_json

index

recipe_json

plate_lock_json

append_locks

state_dimensions

STEP 2B — one variation's prompt, out of the set the LLM returned. Appends the identity and scene locks verbatim, so the authored part varies and the fidelity part cannot.

ArkPromptAt

prompt

report

arkennemasis Prompt At (2B · one variation's prompt)

Name	Type	Default	Description
prompts_json	STRING	[]	The LLM's answer from step 2A.
cell_json	STRING	{}	—
index	INT	00–100000	—
recipe_jsonopt	STRING		Supplies the two locks. Without it the authored prompt is sent alone and identity is not anchored.
plate_lock_jsonopt	STRING		Wire this. The hosted image backend IGNORES the tool's size field, so the ONLY way to control the output shape is to state it in the prompt — and the shape that is wanted is always the locked plate's. Without it the model picks its own canvas and reframes the product.
append_locksopt	BOOLEAN	true	Append the identity and scene locks verbatim. Off only to see what the model wrote on its own.
state_dimensionsopt	BOOLEAN	true	Append the plate's exact pixel size and aspect to every prompt. Requires plate_lock_json.

arkennemasis Prompt At (2B · one variation's prompt)

Inputs (7)

Outputs (3)