We provide weekly real-time updates on the latest diffusion generation models across various fields.
Python
85
94 commits
updated Sep 23, 2026
README
✨ Awesome Generative Models
A high-precision, exact-link catalog of recent generative-model research
for images, video, 3D assets, 3D scenes, and 4D dynamic worlds.
Exact paper / code / project links only — every entry is reviewed before merge.
🧭 Quick Navigation
| Artifact | Focus | Explore by task | Papers |
|---|---|---|---|
| 🖼️ Image 2D | Final output is a single 2D image: text-to-image, controllable generation, safety, personalization, and efficient inference. | text-to-image · controllable-generation · model-efficiency · alignment-safety · personalization | ██████████████ 56 |
| 🎬 Video | Final output is a video: text-to-video, image-to-video, editing, human animation, long video, and surround-view video. | text-to-video · image-to-video · video-editing · human-animation · long-video · autonomous-driving-video | █████████████░ 51 |
| 🧊 3D Object / Asset | Reusable 3D objects, assets, avatars, articulated assets, and part-aware generation. | 3d-shape-generation · image-to-3d · text-to-3d · part-aware-generation · articulated-asset · human-avatar | ██████████░░░░ 40 |
| 🏙️ 3D Scene | Multi-object 3D scene generation, indoor/outdoor scenes, layout-conditioned scenes, and scene editing. | general-scene-generation · layout-to-scene · agentic-scene-generation · indoor · outdoor-urban · single-image-to-scene · scene-editing | █████████░░░░░ 36 |
| 🌍 4D Dynamic Scene / World | Dynamic 3D/4D scenes, world models, simulation, autonomous driving, game worlds, and interactive world modeling. | autonomous-driving · 4d-generation · world-models · simulation · game-worlds · robotics-worlds | ██████░░░░░░░░ 24 |
🌟 Editor's Picks
The highest-signal recent work in each area — newest, most complete, or most widely adopted.
🖼️ Image 2D · see all →
🌟 iFSQ: Improving FSQ for Image Generation with 1 Line of Code
A simple quantization improvement that makes FSQ stronger for high-quality image generation with minimal code changes.
🎬 Video · see all →
🌟 Anti-I2V: Safeguarding your photos from malicious image-to-video generation
A defensive method for protecting photos against misuse by malicious image-to-video generation systems.
🧊 3D Object / Asset · see all →
🌟 AssetFormer: Modular 3D Assets Generation with Autoregressive Transformer
Targets modular assets directly with autoregressive generation over part-aware structures.
🏙️ 3D Scene · see all →
🌟 FlowScene: Style-Consistent Indoor Scene Generation with Multimodal Graph Rectified Flow
A graph-based rectified-flow approach to style-consistent indoor scene generation.
🌍 4D Dynamic Scene / World · see all →
🌟 VectorWorld: Efficient Streaming World Model via Diffusion Flow on Vector Graphs
VectorWorld is a streaming autonomous-driving world model that incrementally generates ego-centric lane-agent vector-graph tiles during rollout. It uses a motion-aware VAE and one-step masked completion on vector graphs to achieve stable, real-time long-horizon simulation. The work is valuable because it targets policy-compatible closed-loop generation instead of offline video-only realism.
vae diffusion flow world-model
📐 How each pipeline works (click to expand)
| Artifact | Input | Core generator | Output |
|---|---|---|---|
| 🖼️ Image 2D | Prompt / control / reference | Diffusion · Flow · AR · VQ latent generator | Single 2D image |
| 🎬 Video | Text / image / pose / camera | Spatiotemporal DiT · diffusion · flow | Video clip / surround-view sequence |
| 🧊 3D Object / Asset | Text / image / multiview prior | 3D latent generator + geometry / texture heads | Reusable 3D object / asset / avatar |
| 🏙️ 3D Scene | Text / layout / room constraints | 3DGS · mesh · occupancy scene generator | Static multi-object 3D scene |
| 🌍 4D Dynamic Scene / World | Observation / action / trajectory | Dynamics rollout · world model · simulator | Interactive world / closed-loop simulation |
🏷️ Explore by Topic
| Topic | Papers |
|---|---|
| autonomous-driving | ██████████████ 18 |
| indoor | ████████░░░░░░ 10 |
| human-avatar | ███████░░░░░░░ 9 |
| gameplay | ████░░░░░░░░░░ 5 |
| outdoor-urban | ██░░░░░░░░░░░░ 3 |
| robotics | █░░░░░░░░░░░░░ 1 |
| tabletop | █░░░░░░░░░░░░░ 1 |
🏢 Explore by Organization
| Organization | Papers |
|---|---|
| Tencent Hunyuan | ██████████████ 9 |
| Independent | ██░░░░░░░░░░░░ 1 |
| Alibaba | ██░░░░░░░░░░░░ 1 |
| ByteDance Seed | ██░░░░░░░░░░░░ 1 |
📬 Daily Candidate Inbox
Each report lists only candidates never surfaced before (a delta, not a re-scan). A day with 0 new items is normal — it means nothing new yet cleared the 2025-01-01+ scope and quality bar.
| Report | Report | Report |
|---|---|---|
| 2026-09-23 | 2026-09-22 | 2026-09-21 |
| 2026-09-20 | 2026-09-19 | 2026-09-18 |
⚙️ Curation Pipeline
Daily harvesting (2025-01-01+ floor, true-novelty dates, quality score, dedupe ledger)
-> metadata/candidates/YYYY-MM-DD/ -> human review -> data/*.jsonl
-> fast validate (schema/scope/dedupe, every push) -> build -> publish
-> weekly deep validate (live links + stars, non-blocking to daily)
🛠️ Build & maintenance commands
# fast: schema, vocab, dedupe, 2025+ scope -- no network, runs on every push
python scripts/validate_data.py --skip-network
# deep: also verifies every exact link + refreshes GitHub star/license cache (weekly)
python scripts/validate_data.py --write-cache
# rebuild README and all generated pages
python scripts/build.py
# refresh the daily candidate inbox
python scripts/fetch_candidates.py --days 14
📖 Recommended Surveys
- Simulating the Real World: A Unified Survey of Multimodal Generative Models
- A Survey of World Models for Autonomous Driving
- The Role of World Models in Shaping Autonomous Driving: A Comprehensive Survey
- Foundation Models in Autonomous Driving: Scenario Generation and Analysis
If this repository helps your research or engineering work, please consider giving it a star.
Contributors
92 commits
2 commits
Similar repos
Languages
Python
100.0%
We provide weekly real-time updates on the latest diffusion generation models across various fields.
Python
85
94 commits
updated Sep 23, 2026
README
✨ Awesome Generative Models
A high-precision, exact-link catalog of recent generative-model research
for images, video, 3D assets, 3D scenes, and 4D dynamic worlds.
Exact paper / code / project links only — every entry is reviewed before merge.
🧭 Quick Navigation
| Artifact | Focus | Explore by task | Papers |
|---|---|---|---|
| 🖼️ Image 2D | Final output is a single 2D image: text-to-image, controllable generation, safety, personalization, and efficient inference. | text-to-image · controllable-generation · model-efficiency · alignment-safety · personalization | ██████████████ 56 |
| 🎬 Video | Final output is a video: text-to-video, image-to-video, editing, human animation, long video, and surround-view video. | text-to-video · image-to-video · video-editing · human-animation · long-video · autonomous-driving-video | █████████████░ 51 |
| 🧊 3D Object / Asset | Reusable 3D objects, assets, avatars, articulated assets, and part-aware generation. | 3d-shape-generation · image-to-3d · text-to-3d · part-aware-generation · articulated-asset · human-avatar | ██████████░░░░ 40 |
| 🏙️ 3D Scene | Multi-object 3D scene generation, indoor/outdoor scenes, layout-conditioned scenes, and scene editing. | general-scene-generation · layout-to-scene · agentic-scene-generation · indoor · outdoor-urban · single-image-to-scene · scene-editing | █████████░░░░░ 36 |
| 🌍 4D Dynamic Scene / World | Dynamic 3D/4D scenes, world models, simulation, autonomous driving, game worlds, and interactive world modeling. | autonomous-driving · 4d-generation · world-models · simulation · game-worlds · robotics-worlds | ██████░░░░░░░░ 24 |
🌟 Editor's Picks
The highest-signal recent work in each area — newest, most complete, or most widely adopted.
🖼️ Image 2D · see all →
🌟 iFSQ: Improving FSQ for Image Generation with 1 Line of Code
A simple quantization improvement that makes FSQ stronger for high-quality image generation with minimal code changes.
🎬 Video · see all →
🌟 Anti-I2V: Safeguarding your photos from malicious image-to-video generation
A defensive method for protecting photos against misuse by malicious image-to-video generation systems.
🧊 3D Object / Asset · see all →
🌟 AssetFormer: Modular 3D Assets Generation with Autoregressive Transformer
Targets modular assets directly with autoregressive generation over part-aware structures.
🏙️ 3D Scene · see all →
🌟 FlowScene: Style-Consistent Indoor Scene Generation with Multimodal Graph Rectified Flow
A graph-based rectified-flow approach to style-consistent indoor scene generation.
🌍 4D Dynamic Scene / World · see all →
🌟 VectorWorld: Efficient Streaming World Model via Diffusion Flow on Vector Graphs
VectorWorld is a streaming autonomous-driving world model that incrementally generates ego-centric lane-agent vector-graph tiles during rollout. It uses a motion-aware VAE and one-step masked completion on vector graphs to achieve stable, real-time long-horizon simulation. The work is valuable because it targets policy-compatible closed-loop generation instead of offline video-only realism.
vae diffusion flow world-model
📐 How each pipeline works (click to expand)
| Artifact | Input | Core generator | Output |
|---|---|---|---|
| 🖼️ Image 2D | Prompt / control / reference | Diffusion · Flow · AR · VQ latent generator | Single 2D image |
| 🎬 Video | Text / image / pose / camera | Spatiotemporal DiT · diffusion · flow | Video clip / surround-view sequence |
| 🧊 3D Object / Asset | Text / image / multiview prior | 3D latent generator + geometry / texture heads | Reusable 3D object / asset / avatar |
| 🏙️ 3D Scene | Text / layout / room constraints | 3DGS · mesh · occupancy scene generator | Static multi-object 3D scene |
| 🌍 4D Dynamic Scene / World | Observation / action / trajectory | Dynamics rollout · world model · simulator | Interactive world / closed-loop simulation |
🏷️ Explore by Topic
| Topic | Papers |
|---|---|
| autonomous-driving | ██████████████ 18 |
| indoor | ████████░░░░░░ 10 |
| human-avatar | ███████░░░░░░░ 9 |
| gameplay | ████░░░░░░░░░░ 5 |
| outdoor-urban | ██░░░░░░░░░░░░ 3 |
| robotics | █░░░░░░░░░░░░░ 1 |
| tabletop | █░░░░░░░░░░░░░ 1 |
🏢 Explore by Organization
| Organization | Papers |
|---|---|
| Tencent Hunyuan | ██████████████ 9 |
| Independent | ██░░░░░░░░░░░░ 1 |
| Alibaba | ██░░░░░░░░░░░░ 1 |
| ByteDance Seed | ██░░░░░░░░░░░░ 1 |
📬 Daily Candidate Inbox
Each report lists only candidates never surfaced before (a delta, not a re-scan). A day with 0 new items is normal — it means nothing new yet cleared the 2025-01-01+ scope and quality bar.
| Report | Report | Report |
|---|---|---|
| 2026-09-23 | 2026-09-22 | 2026-09-21 |
| 2026-09-20 | 2026-09-19 | 2026-09-18 |
⚙️ Curation Pipeline
Daily harvesting (2025-01-01+ floor, true-novelty dates, quality score, dedupe ledger)
-> metadata/candidates/YYYY-MM-DD/ -> human review -> data/*.jsonl
-> fast validate (schema/scope/dedupe, every push) -> build -> publish
-> weekly deep validate (live links + stars, non-blocking to daily)
🛠️ Build & maintenance commands
# fast: schema, vocab, dedupe, 2025+ scope -- no network, runs on every push
python scripts/validate_data.py --skip-network
# deep: also verifies every exact link + refreshes GitHub star/license cache (weekly)
python scripts/validate_data.py --write-cache
# rebuild README and all generated pages
python scripts/build.py
# refresh the daily candidate inbox
python scripts/fetch_candidates.py --days 14
📖 Recommended Surveys
- Simulating the Real World: A Unified Survey of Multimodal Generative Models
- A Survey of World Models for Autonomous Driving
- The Role of World Models in Shaping Autonomous Driving: A Comprehensive Survey
- Foundation Models in Autonomous Driving: Scenario Generation and Analysis
If this repository helps your research or engineering work, please consider giving it a star.
Similar repos
Contributors
92 commits
2 commits
Languages
Python
100.0%