I did not post a lot of seeddance 2.0 sample but here is some. ep1 and ep2 are same consistent character. ep3 is build with 2 individual photo. #danceseed2 @dreamina_ai

Comfyui support for DreamID-V，if you like my node remember to click the star：） original github：github.com/bytedance/DreamID… comfyui node：github.com/TTPlanetPig/Comfy… #comfyui #bytedance #AIvideo #faceswap

Update the TTP_tooset for comfyui. It will bring you a new node to support NEW LTX 2 first and last frame control capability. Workflow are in the sample folder github.com/TTPlanetPig/Comfy…

Mecha Virtual Idol <LINAE-泠奈> Debut Single — "Binary Breath": The Aesthetic of Blue & White Violence. Body: "Cold logic, warm longing." What started as a small experiment into the boundaries of AI in 2025 evolved into my first full-scale Music Video. As LINAE emerged from the blue and white code, I became captivated by the friction between her heavy armor and her delicate soul. About LINAE: "Lin (泠)" stands for the cold, transparent void. "Nae (奈)" represents the resolute yet tender vitality of a girl. She isn't just a 2D illustration; she is stepping into the weight of the real world. The Craft: To avoid the "plastic 3D" look common in AI, I spent weeks fine-tuning custom LoRAs and merging cutting-edge models to achieve that raw contrast between metallic structures and biological skin. This is LINAE’s first breath—and my awakening alongside AI. AI Tech Stack: 🛠️ Google Gemini Pro 🛠️ Jimeng (Dreamina) 3.5 Pro 🛠️ Wan 2.1 Series 🛠️ Qwen-edit 🛠️ Suno AI 🛠️ Flux Custom Trained LoRAs #LINAE #VirtualIdol #AIArt #Cyberpunk #Mecha #GenerativeAI #MusicVideo

Gundam Girl outshape design！ NOW it with great consistency and realistic photo #gundam #mechagirl #RX78

Wan2.2 The quality is significantly improved, tested some human body performance. Amazing @Alibaba_Wan #wan2.2

I didi not realize it could be used in this way

Step1, Pick up a photo you like Step2, Use my PVC Lora for FBL Kontext to convert to image to a high quality figure preview image Step3, Use Tencent Hunyuan 3D model to convert single image to a functional 3D model Step4, Print it, if you can use SLA rather than FDM, you will get much better Surface! @TencentHunyuan @bfl_ml

Kontext PVC lora-real convertion for the realistic photo or comic characters. Super high quality and consistence for output. However, not for free this time as training collection is with very high effort. If you want it, contact me by qq group: 1050777195 or tg group: t.me/ 2Pb1A_T69501NjM1

Thanks KJ for re-build the logic for multitalk in Comfyui now it work fine and amazing @Kijaidesign

Hunyuan 3D 2.5 look much better

Now I put my Modified Start End FramePack logic from my original PR github.com/lllyasviel/FrameP… into comfyui node, I have improved the variability significantly by adjusting the end frame strength and history latent reference. Please use 8s/0.8 end image strength and 0.1 history decay to get better result. github.com/TTPlanetPig/TTP_C… KJ maybe you can check how to optimize the way to allocate the strength control @Kijaidesign

Again Framepack start to end。8s。my felling it is better than close source application. smooth and logical！ #FramePack

Made some experimental again today on frame pack code, now more variable on the start to end frame version. @lvminzhang @TencentHunyuan #FramePack #HunyuanVideo Again Rem 赛高

Just modified @lvminzhang 's framepack and add the start and end frame control!!!! now it can create wonderfull video based on your willingness!!! I have proposed a PR and waiting for approval. the effect as belows: you can decide if you want fast or slow! @TencentHunyuan

Framepack 30s 720P 1s by RTX4090 120s Very stable quality！ Hunyuan reborn！ @TencentHunyuan

No wonder it is much better! Because lvmin Zhang @lvminzhang upgraded the original Hunyuan! Now it is even better than Wan. And with better “human anatomy” which is NSFW By the way, current setting is 540p by modify the code you can run 720p easily! The VRAM consumption is 36G Below video is 720p base @TencentHunyuan

Thanks for @lvminzhang for building such a great tools! Now we don’t need super expensive GPU and high VRam We can easily generate video longer than 5s, even 60s is possible! github.com/lllyasviel/FrameP… With this frame pack program, we have stable image and much faster speed! current the base model is Hunyuan. @TencentHunyuan

tried to use Wan 2.1 to build a dancing clip. I used one image for reference. however created a monster. workflow based on KJ‘s node( ´▽｀)@Kijaidesign @WanxAI_

These pixel cards were created with GPT and a bit of imagination. Each character is unique, every skill has its own twist. From Nugget Hero to RTX4090 Knight, from Snow Dark to Luckin Warrior — they're all citizens of a new world. Creative tools are evolving. In the future, everyone can be a creator. #ChatGPT4o