Blog

TensorArt basics: use a checkpoint as the main image model, add LoRAs to change style or character details, and write a clear prompt to guide the image. Most default settings are fine, while things like aspect ratio, steps, guidance, scheduler, and seed just give more control. Then you log in, click Create, and start generating with your free daily credits.

LTX2 video inpainting setup that edits masked parts of a video, with two versions: a lighter one using only LTX2 + LoRAs, and a heavier one that also uses Qwen for more consistent results. You need a lot of VRAM/RAM, probably cloud GPU, plus several LTX2 model files, LoRAs, and one masking model called SeC-4B. The workflow can use three masking methods: MatAnyone for speed, SeC-4B for better tracking, or SAM3 for prompt-based masking. LTX2 is more flexible than Wan VACE for longer clips, more resolutions, and different FPS, but it is much more resource-hungry.

Single-file ComfyUI video-processing workflow built around Wan VACE that combines model loading, automatic object segmentation, masking, generation, and optional upscale/interpolation. You download the workflow JSONs plus a set of required models, place each one in the exact ComfyUI folder expected by the nodes, make sure the MatAnyone custom node is installed, then load the workflow and point it at your input video.

Quick-start guide for running Wan 2.1 through Pinokio instead of setting it up manually. It says to install Pinokio, install Wan 2.1 from the built-in browser, start the image-to-video 480p model, add compatible LoRA files into the loras_i2v folder, refresh the app so they appear, and then generate videos with prompts. It also gives a performance estimate: on an 8 GB VRAM GPU with 32 GB RAM, a 5-second clip takes about 20 minutes, so shorter clips are recommended unless you have a stronger GPU.

You need ComfyUI, ComfyUI Manager, the workflow file, and the Realistic Vision Inpaint model placed in ComfyUI/models/checkpoints/. It runs best with 6 GB+ VRAM, but 4 GB can work with low-VRAM/FP16 settings. Load the workflow, install any missing nodes, add and resize your image, use the Florence2 mask step to detect the area you want edited by describing it in text, set positive and negative prompts, then run it. On the first run, Florence and SAM2 download automatically.

This is a setup guide for a ComfyUI workflow that uses FLUX Kontext models to remove clothing in images. It tells you which model files to download, where to place them in the ComfyUI folders, which version to choose based on your GPU VRAM, and which workflow JSON to import. Low-end GPUs are told to use quantized GGUF files and an extra LoRA/plugin. The basic flow is: install ComfyUI, download the required encoders/VAE/model files, put them in the right directories, install ComfyUI-GGUF if needed, load the matching workflow, upload an image, adjust the prompt, and generate.