[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-83239":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":13,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":12,"stars7d":15,"stars30d":15,"stars90d":14,"forks30d":14,"starsTrendScore":16,"compositeScore":17,"rankGlobal":9,"rankLanguage":9,"license":9,"archived":18,"fork":18,"defaultBranch":19,"hasWiki":18,"hasPages":20,"topics":21,"createdAt":9,"pushedAt":9,"updatedAt":22,"readmeContent":23,"aiSummary":9,"trendingCount":14,"starSnapshotCount":14,"syncStatus":12,"lastSyncTime":24,"discoverSource":25},83239,"Echo-Infinity","Echo-Team-Joy-Future-Academy-JD\u002FEcho-Infinity","Echo-Team-Joy-Future-Academy-JD","Official repo for paper \"Echo-Infinity: Learnable Evolving Memory for Real-Time Infinite Video Generation\"",null,"Python",84,2,1,0,22,13,1.43,false,"main",true,[],"2026-06-12 02:04:32","\u003Cdiv align=\"center\">\n\n# Echo-Infinity\n### Learnable Evolving Memory for Real-Time Infinite Video Generation\n\n\u003Ch3 align=\"center\">\u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.04527\">Paper\u003C\u002Fa> | \u003Ca href=\"https:\u002F\u002Fecho-team-joy-future-academy-jd.github.io\u002FEcho-Infinity\u002F\">Website\u003C\u002Fa> | \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002FEcho-Team\u002FEcho-Infinity\">Models\u003C\u002Fa> | \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FEcho-Team-Joy-Future-Academy-JD\u002FEcho-Infinity\">Code\u003C\u002Fa>\u003C\u002Fh3>\n\n\u003C\u002Fdiv>\n\n-----\n\nEcho-Infinity demonstrates hour-scale and real-time video generation with a learnable memory to filter, abstract, and compress any-length history at constant cost, suggesting a practical path toward infinite video generation.\n\n-----\n\n\u003Ctable align=\"center\">\n\u003Ctr>\n\u003Ctd align=\"center\" width=\"50%\">\n  \u003Ca href=\"https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=YR7G_yJs8WM\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.youtube.com\u002Fvi\u002FYR7G_yJs8WM\u002Fhqdefault.jpg\" alt=\"24h Demo — Part 1 \u002F 2\" width=\"100%\"\u002F>\n    \u003Cbr\u002F>\n    \u003Csub>\u003Cb>24h Demo — Part 1 \u002F 2\u003C\u002Fb>\u003C\u002Fsub>\n  \u003C\u002Fa>\n\u003C\u002Ftd>\n\u003Ctd align=\"center\" width=\"50%\">\n  \u003Ca href=\"https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=kF2Nksvijb8\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.youtube.com\u002Fvi\u002FkF2Nksvijb8\u002Fhqdefault.jpg\" alt=\"24h Demo — Part 2 \u002F 2\" width=\"100%\"\u002F>\n    \u003Cbr\u002F>\n    \u003Csub>\u003Cb>24h Demo — Part 2 \u002F 2\u003C\u002Fb>\u003C\u002Fsub>\n  \u003C\u002Fa>\n\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftable>\n\n\u003Cp align=\"center\">\n\u003Csub>\u003Ci>Note: Each 24-hour demo is too large to host inline, so it is only viewable via YouTube. Each clip is split into two consecutive 12-hour parts due to YouTube's per-video duration limit, and visual quality has been moderately compressed for upload bandwidth efficiency.\u003C\u002Fi>\u003C\u002Fsub>\n\u003C\u002Fp>\n\n\n## 🔥 News\n- **2026.6.03**: The [paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.04527), [project page](https:\u002F\u002Fecho-team-joy-future-academy-jd.github.io\u002FEcho-Infinity\u002F), [model](https:\u002F\u002Fhuggingface.co\u002FEcho-Team\u002FEcho-Infinity), and [code](https:\u002F\u002Fgithub.com\u002FEcho-Team-Joy-Future-Academy-JD\u002FEcho-Infinity) are released.\n\n\n## Quick Start\n\n### Installation\n\n```bash\nconda create -n echo_infinity python=3.10 -y\nconda activate echo_infinity\n\ncd Echo-Infinity\npip install -r requirements.txt\npip install flash-attn --no-build-isolation\npython setup.py develop\n```\n\n### Download Checkpoints\n\n```bash\n# Wan2.1 base models (teacher \u002F student backbones)\nhf download Wan-AI\u002FWan2.1-T2V-1.3B --local-dir wan_models\u002FWan2.1-T2V-1.3B\nhf download Wan-AI\u002FWan2.1-T2V-14B  --local-dir wan_models\u002FWan2.1-T2V-14B\n\n# Stage-2 (Causal ODE) init from upstream Causal-Forcing\nhf download zhuhz22\u002FCausal-Forcing chunkwise\u002Fcausal_forcing.pt --local-dir checkpoints\n\n# Echo-Infinity Stage-1 (init) and Stage-2 (long) checkpoints\nhf download Echo-Team\u002FEcho-Infinity echo_infinity.pt       --local-dir checkpoints\nhf download Echo-Team\u002FEcho-Infinity echo_infinity-long.pt  --local-dir checkpoints\n```\n\n### CLI Inference\n\nAll commands assume cwd = `Echo-Infinity\u002F`.\n\n**5s — short video** (single-prompt, EMA on):\n```bash\nCUDA_VISIBLE_DEVICES=0 python inference\u002Finference.py \\\n    --config_path configs\u002Fecho_infinity_inference_std.yaml \\\n    --use_ema \\\n    --output_folder output\u002F5s \\\n    --seed 0\n```\n\n**30s — mid-length video** (single-prompt):\n```bash\nCUDA_VISIBLE_DEVICES=0 python inference\u002Finference.py \\\n    --config_path configs\u002Fecho_infinity-long_inference.yaml \\\n    --output_folder output\u002F30s \\\n    --seed 0\n```\n\n**240s — long video** (single-prompt):\n```bash\nCUDA_VISIBLE_DEVICES=0 python inference\u002Finference.py \\\n    --config_path configs\u002Fecho_infinity-long_inference_240s.yaml \\\n    --output_folder output\u002F240s \\\n    --seed 0\n```\n\n**60s interactive** (multi-prompt switching within one video):\n```bash\nCUDA_VISIBLE_DEVICES=0 python inference\u002Finteractive_inference.py \\\n    --config_path configs\u002Fecho_infinity-long_interactive.yaml \\\n    --output_folder output\u002F60s_interactive \\\n    --seed 1\n```\n\n**1h — hour-level video** (streaming decode):\n```bash\nbash inference\u002Fstream_long\u002Frun_1h.sh\n```\n\n**24h — full-day video** (streaming decode):\n```bash\nbash inference\u002Fstream_long\u002Frun_24h.sh\n```\n\nPrompts are under `inference\u002Fprompts\u002Fdemo_*.txt` and switch-prompts at `inference\u002Fprompts\u002Fdemo_60s_interactive.jsonl`. Override with `--data_path your_prompts.txt`.\n\n\n## Training\n\nThe pipeline has two stages of DMD training. Both are launched on 4 nodes × 8 GPUs by default (`gradient_accumulation_steps=2`, effective batch size 64). Override the launch topology via `MASTER_ADDR`, `NODE_IP_*`, `NNODES`, and `NPROC_PER_NODE` environment variables (e.g. `NNODES=1 NPROC_PER_NODE=8 bash scripts\u002Ftrain_echo_infinity_init.sh` for single-node training).\n\nWeights & Biases logging is **off by default**. To enable it, set `USE_WANDB=1` and fill in `wandb_key` \u002F `wandb_entity` in the corresponding config (`configs\u002Fecho_infinity.yaml`, `configs\u002Fecho_infinity-long.yaml`).\n\n### Stage 1 — Init\n\n```bash\nbash scripts\u002Ftrain_echo_infinity_init.sh\n```\n\nOutput: `logs\u002Fecho_infinity\u002Fcheckpoint_model_000400\u002Fmodel.pt`. To reuse it as the Stage-2 init or for inference, copy it to `checkpoints\u002Fecho_infinity.pt` (the path the configs expect) or pass `--checkpoint_path`.\n\n### Stage 2 — Long-Video Tuning\n\n```bash\nbash scripts\u002Ftrain_echo_infinity_long.sh\n```\n\nOutput: `logs\u002Fecho_infinity-long\u002Fcheckpoint_model_003200\u002Fmodel.pt`. To reuse it for the long-form inference above, copy it to `checkpoints\u002Fecho_infinity-long.pt` (the `lora_ckpt` path the configs expect) or pass `--lora_ckpt`.\n\nTraining data (download from the same HF repo as the model weights):\n\n```bash\nhf download Echo-Team\u002FEcho-Infinity vidprom_filtered_extended.txt        --local-dir prompts\nhf download Echo-Team\u002FEcho-Infinity vidprom_filtered_extended_switch.txt --local-dir prompts\n```\n\n- `prompts\u002Fvidprom_filtered_extended.txt`        — base prompts for streaming training\n- `prompts\u002Fvidprom_filtered_extended_switch.txt` — prompt-switch pairs for interactive training\n\n\n## Acknowledgements\n\nThis codebase builds on the open-source implementations of:\n- [Wan2.1 (Wan-Video)](https:\u002F\u002Fgithub.com\u002FWan-Video\u002FWan2.1)\n- [Causal-Forcing (thu-ml)](https:\u002F\u002Fgithub.com\u002Fthu-ml\u002FCausal-Forcing)\n- [LongLive (NVlabs)](https:\u002F\u002Fgithub.com\u002FNVlabs\u002FLongLive)\n- [Self-Forcing (guandeh17)](https:\u002F\u002Fgithub.com\u002Fguandeh17\u002FSelf-Forcing)\n\n\n## References\n\n```\n@article{bian2026echoinfinity,\n  title={Echo-Infinity: Learnable Evolving Memory for Real-Time Infinite Video Generation},\n  author={Bian, Yuxuan and Xue, Zeyue and Zhang, Songchun and Zhang, Shiyi and Jin, Weiyang and Li, Yaowei and Zhuang, Junhao and Li, Haoran and Huang, Jie and Huang, Haoyang and Duan, Nan and Xu, Qiang},\n  journal={arXiv preprint arXiv:2606.04527},\n  year={2026}\n}\n```\n","2026-06-11 04:10:31","CREATED_QUERY"]