Tasks and inputs
SenseNova-U1.5-8B-MoT is listed for Any-to-any, Image-to-image, Text-to-image. It accepts text, image and produces text, image. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 18, 2026
SenseNova-U1.5-8B-MoT is SenseTime's open unified model for native 4K image generation and instruction-based image editing, Apache-2.0 licensed.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: No API pricing stated; open weights for self-hosting
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Posters break most image models. Small Chinese captions turn to noise, the hierarchy collapses, and a second round of edits quietly changes the face you asked it to keep. SenseNova-U1.5-8B-MoT is SenseTime's August 2026 answer to exactly those failures, and it ships as open weights.
It is a native unified multimodal checkpoint built on SenseTime's NEO-unify design: one model reads text and images and generates images, rather than a language model bolted to a separate diffusion decoder. U1.5 rebuilds the patchify layers, the data mix, the task formulation, prompt enhancement and post-training of the earlier U1.
| Checkpoint | What it is |
|---|---|
sensenova/SenseNova-U1.5-8B-MoT |
Main RL-tuned release |
SenseNova-U1.5-8B-MoT-SFT |
Supervised stage only |
SenseNova-U1.5-8B-MoT-TailSteer |
4 November 2026, steadier multi-turn editing |
| U1.5 LoRAs | 8-step adapter (August) and an 8-step V2 (24 September) |
The weights total roughly 17.5B parameters on disk, despite the "8B" in the name. Reference code runs on Python 3.11, PyTorch 2.8 and CUDA 12.8. The Hub repository is dated 19 August while the project's own release log says 20 August; either way, it is the same checkpoint.
SenseTime is candid about weak spots. Colours can oversaturate (lowering cfg_scale helps). Dense, tiny or mixed-language text still slips. Exact counts in tight layouts, small hands and faces, and long multi-reference editing sessions can drift. That last point is why TailSteer exists.
Take twenty briefs your designers actually received last month. Generate each at 2048×2048, then run three rounds of edits that must preserve a face or a logo, and score drift per round. Compare it against a hosted option such as Qwen-Image-2.1-Pro and the rest of the text-to-image directory. If multi-turn stability is your bottleneck, rerun the editing rounds with TailSteer before you decide. A free browser playground, SenseNova-Studio, lets non-engineers judge output before anyone provisions a GPU.
Compare the model on the work you need to complete.
SenseNova-U1.5-8B-MoT is listed for Any-to-any, Image-to-image, Text-to-image. It accepts text, image and produces text, image. Use the publisher’s input format and preprocessing for your chosen deployment.
Choose the output size and quality your design needs. Include drafts and revisions when budgeting an accepted image.
The weights are downloadable under Apache-2.0. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Liquid AI
LFM3-Omni-2B is Liquid AI's open any-to-any model: image, video, audio and text in; text and speech out, with multimodal tool calls on-device.
NVIDIA
NLD-Uni (Nemotron-Labs-Diffusion-Uni-8B) is NVIDIA's open diffusion-language model that both understands images and generates them, for research.
NVIDIA
PixelUMM is NVIDIA's encoder-free 15B unified model that understands and generates text, images and video directly in pixel space. Non-commercial.
shanghai-ai-lab
InternLumina-U2 is Shanghai AI Lab's masked-diffusion MoE that understands images, video and 3D and generates and edits images with 8-codebook visual tokens.
Tencent
Hy-Embodied-RxBrain-1.0 is Tencent's 6.2B embodied model that reasons in text and imagines goal images, for planning, world-state prediction and video QA.
google/gemma-4-E2B-it is a source-linked Hugging Face repository assigned to Any-to-any. Its recorded task interface accepts text, image, audio, video and produces text, image, audio, video.