AI Image2026
A distilled LoRA adapter for Qwen-Image-2.1 that runs text-to-image generation and instruction-driven image editing in a few transformer passes (shipped as a 6-step r256 LoRA). Samples with a fixed sigma schedule, no classifier-free guidance; non-commercial research license.
AI Video2026
Replaces a selected person in a source video with a character from a reference image via a MiniMax H3 LoRA adapter, aiming to preserve scene, camera, and background. Trained for 1,000 updates; intended for Ref2VA runtimes and ComfyUI. Short (≈4–5s) continuous shots work best; distributed under the MiniMax H3 Community License.
AI Video2026
Turns a single photo into a geometry-consistent, frozen-time 360° camera orbit that returns to the exact start frame. Implemented as a LoRA for MiniMax‑H3 FL2VA — use identical first+last keyframes to produce seamless orbit clips; trained on a small human-centric square orbit dataset, so results are domain-limited.
Demonstrates that pretrained transformers typically use only ~1–3 lines of depth to follow reference chains, and that a task‑trained rank‑8 LoRA applied at one early layer (with all other weights frozen) can extend reference‑following to dozens or hundreds of lines while adding only a few ten‑thousand parameters.
AI Model2026
Delivers fast, calibrated 'System 1' decisions over text and images (yes/no, 2–256 choices, 0–5 scores) in one forward pass, plus full 'System 2' Qwen3.8-27B multimodal reasoning. Highlights: zero-shot vision judging, 256K-token context, vLLM decision API, Apache-2.0.
AI Model2026
Provides calibrated probabilistic decisions (yes/no, 2–256 choice, 0–5 score) in one forward pass, with an optional adaptive-thinking mode that invokes Gemma‑4 when System 1 is uncertain; supports text+image, 256K context and vLLM serving, but adaptive thinking is much slower.
Adapts LLM agents online by self-distilling verified execution trajectories into persistent LoRA weights during deployment to improve success and efficiency on long‑horizon tasks. Uses a frozen stable copy as a privileged teacher to predict hindsight next‑token distributions and filters invalid-action turns so experience consolidates without external solutions or memory retrieval.
AI Model2026
Returns calibrated probabilities for yes/no and multi-choice decisions using a two-stage System 1 (fast classifier) and System 2 (Gemma‑4 reasoning) pipeline; this NVFP4 release quantizes the 3,840 routed MoE experts to 4-bit so the model fits ≈17–18 GB of GPU memory while other weights remain bf16.
AI Model2026
Generates calibrated decisions (choice/score/noul) from a prompt and optional images, up to 256 options. Routed MoE experts are quantized to NVIDIA NVFP4 to reduce weights to ~25.6 GB and GPU memory to ~23.3 GiB; provided as a vLLM-compatible System 1 decision model with a /v1/decide endpoint.
AI Image2026
Open-weight 5.6B FLUX.3 image model for text-to-image and multi-reference image editing with pixel-accurate bounding-box layout control. Ships distilled (fast, low-step) and base (fine-tuneable) checkpoints and runs on consumer GPUs down to an RTX 3090.
AI Image2026
Runs the FLUX 3 Image Klein single-file checkpoint in ComfyUI for text-to-image generation and instruction-driven image editing with up to ten reference images, bounding-box layout control, and distilled/base variants; requires a prompt upsampler.
AI Model2026
Distills MOSS-Image-SFT from a 50-step, CFG-guided process into image generation with eight network evaluations and no CFG. The checkpoint is still a work in progress.