dots.tts 环境诊断 · T8star-Aix

When dots.tts fails in a way that smells environmental - "the loader says missing model but I downloaded it," a transformer version error at import, a CUDA mismatch - T8_DotsTTS_Environment is the first node to run. It inspects your Python, the pack's dependencies, your CUDA state, and the current model cache, and returns everything as a single JSON report. It's read-only by design: it checks, it doesn't install, doesn't fix, doesn't touch anything. Think of it as python -m pip check plus nvidia-smi plus a model-folder listing, bundled into one node output.

The node takes no inputs at all. Execute it and you get one output, environment_report, a JSON string covering the things that most commonly break this pack. That includes your Python version - a meaningful check because dots.tts v0.3.1 officially supports Python 3.10–3.12 and not 3.13 - plus the presence and versions of the key dependencies, whether CUDA is visible, and what the current model cache state looks like. The T8_DotsTTS_Diagnostics node builds on the same inspection and packages it into a support ZIP; this one just shows you the raw findings.

Where it earns its keep is diagnosis speed. The pack's own README documents a real, known failure pattern: dots.tts v0.3.1 requires transformers>=4.57,<5, and if your environment reports a too-old version, the fix is specific - back up, then python -m pip install "transformers>=4.57,<5". But before you go upgrading shared core libraries in your ComfyUI environment (which can break other nodes - see the ecosystem notes on dependency collisions), you want to confirm the report actually says that. That's what this node is for: it turns "something's wrong somewhere" into "here is the exact thing, in JSON, verified."

Practical notes

• It's free to run - no model load, no GPU work, instant. Run it any time before filing an issue or blaming your setup.
• The output is a JSON string. Wire it to a text preview node to read it comfortably.
• Because it's strictly read-only, it's safe to run mid-workflow without side effects. There's no reason not to have it in a debugging graph.

Installation is just the pack itself (ComfyUI Manager search dots-tts-t8, or clone + pip install -r requirements.txt). It doesn't need a model downloaded - it'll happily tell you that your models are missing, which is frequently the actual diagnosis.