Fix prompt validation: declare lm_top_k/repainting_start/repainting_end as STRING to survive stale empty-string widget values

[Copilot]

PR #11's VALIDATE_INPUTS fix was based on a wrong assumption. ComfyUI's execution.py runs int(val) / float(val) unconditionally in an else branch (lines 901–936) before VALIDATE_INPUTS is ever called — so listing fields there bypasses only range (min/max) checking, not type coercion. int('') and float('') still raise, errors are recorded, and the prompt fails even when VALIDATE_INPUTS returns True.

Fix

Declare lm_top_k, repainting_start, and repainting_end as STRING instead of INT/FLOAT in INPUT_TYPES. ComfyUI then runs str('') (always succeeds). The existing _coerce_int / _coerce_float helpers in generate() convert empty strings to numeric defaults at runtime.

# Before — ComfyUI runs int('') / float('') → ValueError before generate() is reached
"lm_top_k":       ("INT",   {"default": 0,    "min": 0, "max": 1000}),
"repainting_start": ("FLOAT", {"default": -1.0, "min": -1.0, "max": 600.0}),
"repainting_end":   ("FLOAT", {"default": -1.0, "min": -1.0, "max": 600.0}),

# After — ComfyUI runs str('') → '' → _coerce_int/float handles it
"lm_top_k":       ("STRING", {"default": "0"}),
"repainting_start": ("STRING", {"default": "-1.0"}),
"repainting_end":   ("STRING", {"default": "-1.0"}),

Files changed

• nodes.py — Change the three fields to STRING; update generate() signatures to Union[str, int] / Union[str, float]; correct VALIDATE_INPUTS docstring (it does not bypass type coercion).
• workflow-examples/*.json (all 4) — Widget positions 15, 19, 20 updated from JSON numbers to JSON strings ("0", "-1.0") to match the new type declarations.
• tests/test_nodes.py — Three new tests assert STRING type for the changed fields; updated TestValidateInputs docstring.
• tests/test_workflows.py — Workflow tests for positions 15/19/20 now check parseability (int(str(val)) / float(str(val))) instead of isinstance(…, int/float).

---

💬 Send tasks to Copilot coding agent from Slack and Teams to turn conversations into code. Copilot posts an update in your thread when it's finished.