(deforum) Base Parameters

If you open any Deforum Nodes example workflow, this is the first node on the left. (deforum) Base Parameters is the top of the parameter chain - it sets the canvas size, how seeds behave frame to frame, which sampler and scheduler drive every step, and whether Deforum's prompt weighting is on. It outputs a deforum_data dict, and every other parameters node in the pack (ColorMatch, Cadence, Animation, and so on) is built to merge its own settings into that same dict. Chain them in a row and you've assembled the full Deforum configuration blob that the sampler nodes consume.

How it works

The pack's parameter nodes all inherit from a shared base that just does deforum_data.update(**kwargs): take the incoming dict (or start a fresh one), add your settings on top, pass it along. So this node is less a computation and more the first entry in a config table. The important part is what it puts in that table - because those values get read on every single frame of the animation.

The inputs that matter

• width and height - canvas size in pixels, default 512×512. This is the resolution every frame is generated at. Bigger looks sharper but multiplies per-frame cost; for long animations people usually stay modest here and upscale after.
• seed_schedule - default 0: (-1), which is Deforum for "random seed at frame 0." A -1 means "roll a fresh seed"; put a real number there to lock the base seed. This is a keyframe-style schedule string, so 0: (123), 50: (456) changes seed halfway through.
• seed_behavior - fixed or random. Fixed keeps reusing the scheduled seed (good for iteration and for img2img continuity); random rolls a new one each frame, which you'll want for certain noise-driven effects.
• sampler_name and scheduler - the full ComfyUI sampler/scheduler lists. euler + karras is the boring, reliable default that most Deforum users camp on; you can hot-swap for a different look since it's just a dropdown.
• prompt_weighting / normalize_prompt_weights - Deforum's old-school prompt weighting ((word:1.3) syntax), inherited from the A1111 extension days. If you're writing plain prompts with no weights, leave them on - harmless and safe. If you see weird prompt behavior, flipping log_weighted_subprompts on prints the parsed weights to the console so you can see exactly what the CLIP encoder received.

Installing it

Part of the Deforum Nodes pack by XmYx. Install through ComfyUI Manager (search "Deforum Nodes") or:

cd ComfyUI/custom_nodes
git clone https://github.com/XmYx/deforum-comfy-nodes.git

Restart ComfyUI afterward. The first launch triggers install.py, which pip-installs the pack's heavy stack (librosa, opencv-contrib-python, moviepy, pandas, numpy<2.0.0, plus the deforum-studio backend) - expect the first boot to take a while. The README insists on Python 3.10.

Where people get burned

The classic mistake is treating this like a ComfyUI "Empty Latent" node and never wiring its deforum_data output anywhere. If the sampler doesn't receive the dict, your animation runs with defaults instead of your settings - and defaults at 512×512 with a random seed and euler are fine, but your careful config is silently ignored. Follow the data flow: Base Parameters → (more params nodes) → your sampler node. Also, remember this is a shared config, not per-image settings - changing width mid-project changes every frame of the animation.