PrismStream
Flow-matching generation with region control
A text-to-image stack built around region conditioning, so a prompt can say where something belongs rather than only what it is.
- Category
- Research
- Status
- Active
- Language
- Python
- Licence
- Apache-2.0
- Stars
- 2.1k
- Since
- 2026
PrismStream started as a conditioning experiment inside MochiCore and outgrew the notebook it was living in.
Region conditioning
Most control mechanisms bolt onto a finished model: inpainting, masked guidance, attention surgery at sample time. PrismStream trains with regions in the conditioning path from the start, which means the model learns that a mask is a statement about composition rather than a repair instruction.
Sampling
The flow-matching formulation buys us short schedules. Twelve steps produces output we are willing to publish; four is usable for previews.
Status
Active, and the part of the lab moving fastest. Expect the API to change.
References
- [1]Flow Matching for Generative Modeling. Lipman et al., ICLR 2023.
Writing
Regions, not repairs
Inpainting treats a mask as damage. PrismStream treats it as composition. Notes from six weeks of getting that distinction to survive training.
Related projects
BluzenTagger
Multi-label tagging for illustration
A tagger over a 12k-tag vocabulary that holds up on the long tail, where general-purpose vision models quietly collapse into the fifty most common labels.
- Language
- Python
- Stars
- 1.4k
- Licence
- Apache-2.0
- Category
- Research
SakuraDetect
Small detectors for illustration
Face, head, person and hand detection tuned for drawn images, small enough to run on a laptop CPU in real time.
- Language
- Python
- Stars
- 820
- Licence
- MIT
- Category
- Research
TagFlow
Prompt expansion that respects your tags
A small language model that turns a handful of tags into a full prompt without inventing subject matter you did not ask for.
- Language
- Python
- Stars
- 310
- Licence
- Apache-2.0
- Category
- Research