Models.md

June 8, 2026 · View on GitHub

Pre-quantized checkpoints were recommended for most architectures, but on-the-fly quantization with ConvRot is better in all cases. However, ConvRot is also a little slower, so these prequantized models are still useful. Avoid using INT8 Tensorwise models.

Shoutout to vistralis for these:

ModelLink
FLUX.2-klein-base-9bDownload
FLUX.2-klein-base-4bDownload
FLUX.2-klein-9bDownload
FLUX.2-klein-4bDownload

ConvRot:

ModelLink
Ideogram-4Download
LTX2.3 10ErosDownload
Sulphur2 Base (LTX2.3 Finetune)Download
Chroma1 HDDownload
Ernie ImageDownload
Anima Preview 3Download
Flux 2 Klein BaseDownload
LTX2.3 DevDownload
LTX2.3 DistilledDownload
WAN 2.2Download

Outdated int8 models:

ModelLink
Chroma1-HD²Download
Z-Image-Base¹Download
Z-Image-Turbo²Download
AnimaDownload

¹Z-Image Base weights have been Deprecated in favor of Convrot OTF, which is higher quality.

²Tensorwise models are worse than on the fly quantization since we switched to row-wise INT8