BEYOND TEXT MODELS
Image and audio tools open up different possibilities, but they need different hardware checks. Start with a specific workflow and its published requirements.
ComfyUI connects models and processing steps into visual workflows. The application and the image model are separate choices, so installing the app does not establish which workflows fit.
Whisper turns speech into text. Its official repository publishes approximate GPU memory requirements by model size, which are useful starting references for that implementation.
| Model | Published approximate VRAM | What to know |
|---|---|---|
| tiny / base | ~1 GB | Start small to check your recording workflow. |
| small | ~2 GB | Test accuracy on your language and recording quality. |
| medium | ~5 GB | Allow for runtime differences and other GPU use. |
| large | ~10 GB | Check exact variant and implementation. |
| turbo | ~6 GB | Optimised transcription variant; check task support. |
Source: OpenAI Whisper model table, checked 17 September 2026. GB labels are reproduced as published, not converted into our calculator's GiB. These are not speed estimates. Different implementations, batching and precision can change memory use.
Resolution, batch size, the model, precision, text encoders, decoding and extra workflow components all affect the peak. We have not yet validated an image-generation memory calculator, so we do not reuse the text-model formula here.
For a useful comparison, record the exact workflow, model files, software version, GPU, resolution and batch size alongside measured peak memory. That is more informative than an unsupported “runs on 8 GB” claim.