Free in this ecosystem means three different things, and knowing which one you are installing is most of the decision. An open-source addon is free like a workshop is free: the tools cost nothing, the electricity does not. A free tier is the demo of a paid product. A free wrapper bills cloud credits in the background.
With that lens, here is what is genuinely usable today, organised by the job rather than by the hype cycle.
Generate textures and images
Dream Textures runs Stable Diffusion inside Blender: generate, tile and repaint textures straight into an image editor window. Open source, works with decent consumer GPUs, and the tile mode is the feature materials people actually use. Development has slowed in recent years, so check it against your Blender version before building a pipeline around it.
AI Render takes the opposite direction: it feeds your viewport blockout through a diffusion model with ControlNet, so a grey primitives scene comes back as a painted concept. Free and open source, same appetite for VRAM, and a genuinely useful tool for mood passes before modelling detail.
Command Blender in plain English
BlenderGPT turns typed sentences into Blender Python and runs them: build a table, set up lighting, rename every object in the scene. The addon is free; the intelligence is your own OpenAI API key. Keep autosave on and read the generated script before running it - executed code can empty a scene as easily as populate one.
The same idea exists over the Model Context Protocol, the standard that lets a general assistant drive Blender through structured calls rather than raw code. Different contract, same rule: review before execute.
Capture motion from video
Markerless capture is the second lane where free tiers are real:Rokoko and Move AI both film a person and export animation toward Blender, and both cap or throttle their free plans. The workflow that turns a capture into a clean action for a rig is its own craft - format, retarget, cleanup - which is why it has a page of its own here.
The empty shelf: text-to-3D
One honest gap. There is no strong free addon that generates production-ready 3D models on a local machine, and the reason is arithmetic rather than conspiracy: the models are large, the output still needs retopology and materials, and most of that finishing work happens inside Blender anyway. Web generators offer limited free tiers for the generation step itself, and in-scene generation is where paid assistants live. Free covers the ring around the mesh better than it covers the mesh.
Install once, stay safe
01 - Use the official channels. Blender 4.2 and later has the Extensions platform in Preferences; older addons install from disk with a zip downloaded from the project repository.
02 - Disable what idles. Every addon loads at startup; a folder of experiments quietly taxes every session. Keep the two you use, archive the rest.
03 - Watch VRAM before anything else. Diffusion addons fail at the GPU, not in the logic - close the browser with forty tabs and retry before filing a bug.
04 - Archive the source zip. When a project goes quiet, the zip you saved is the difference between a tool and a memory.
The cost maths of free
Local generation costs nothing per image and meaningful money per hour of GPU time and electricity; hosted tools are the reverse. A mixed stack is usually right: open source for bulk iteration, hosting or a paid seat for the hero shot. Measure once and the choice stops being philosophical.
Free is a fine place to start in this corner of the ecosystem, and a surprisingly capable one: textures, control and capture are all covered for the price of hardware you already own. The parts free does not cover yet - mesh quality and the decisions inside it - are the parts that were always the real work.