That's what makes iNodes special: one universal add-on. Every node works the same way, whatever the model and the software, and that goes for your Comfy workflows too. Once you know how to use one node, you know how to use them all: only the settings and the inputs change.
Labels stay in English in your software, whatever the app's language.
Our example is Seedance 2.5 (Reference to Video), one of the most complete nodes, with a video shot and an image as inputs.
In your software
- The inputs: wire your images and videos into the node.
- The result: it arrives as a MediaIn, under the node.
- Prompt: what you want to get. Edit opens a larger editor.
- Generate: starts the generation. Cancel stops it.
- Cost Est.: works out the cost with your current settings.
- Model info: what this model does, and its settings.
- Version: the generation the node shows.
- Skip import generation: stops creating MediaIn nodes.
- Frame Start, Frame End: the part of the video sent to the model. It's the key setting for a video input, see below.
- The inputs: wire your images and videos into the node.
- The result: it arrives as a Read, under the node.
- Prompt: what you want to get. Edit opens a larger editor.
- Generate: starts the generation. Cancel stops it.
- Cost Est.: works out the cost with your current settings.
- Model info: what this model does, and its settings.
- Version: the generation the node shows.
- Skip import generation: stops creating Read nodes.
- Frame Start, Frame End: the part of the video sent to the model. It's the key setting for a video input, see below.
- The inputs: wire your images and videos into the node.
- The result: it arrives as a clip, in the Batch.
- Edit Prompt: what you want to get.
- Generate: starts the generation. Cancel stops it.
- Cost Est.: works out the cost with your current settings.
- Model info: what this model does, and its settings.
- Version: the generation the node shows.
- Skip import: stops creating clips in the Batch.
- Deliver as Open Clip: each generation becomes a version of the same clip.
- Frame Start, Frame End: the part of the video sent to the model. It's the key setting for a video input, see below.
An iNodes node has a lot of inputs in Flame, and that's normal: Flame adds a matte input to each of them, which doubles them. iNodes doesn't use these mattes, so there's no need to plug anything into them.
With Deliver as Open Clip, each generation becomes a new version of the same clip, instead of a new clip in the Batch. The Flame page tells you more.
- The inputs: at the bottom of the node, one socket per image or video.
- The result: it shows above the node, and also arrives in an Image node, below.
- Prompt: what you want to get. Edit opens a larger editor.
- Generate: starts the generation. Cancel stops it.
- Cost Est.: works out the cost with your current settings.
- The “?” icon: what this model does, and its settings.
- Version: the generation the node shows.
- Skip import generation: stops creating Image nodes.
- The inputs: on the left of the node, one socket per image or video.
- The result: here, it shows right inside the node.
- Prompt: what you want to get.
- Generate: starts the generation. Cancel stops it.
- Cost Est.: works out the cost with your current settings.
- Info: what this model does, and its settings.
- Version: the generation the node shows.
- Skip import generation: stops creating a node for the result.
- Frame Start, Frame End: the part of the video sent to the model. It's the key setting for a video input, see below.
Click an image to open it full size.
The inputs
Each model has its own inputs. Wire the ones you need. The model's Model info page tells you what it accepts.
- Video 1 to 6 and Image 1 to 6: the references of a Reference model. Mention them in the prompt, in input order (@Video1, @Image1, depending on the model).
- Source and Last Frame: on an image-to-video model, the first and the last image. Last Frame is usually optional, but a few models require it.
Frame Start and Frame End
As soon as a video goes into a node, this is the most important setting. Frame Start and Frame End decide which part of your shot the model receives, from its first frame to its last. Nothing else is sent. Set them wrong and the model works on the wrong moment of the shot, however good your prompt is.
Before you start a generation, check three things:
- The right part. Frame Start is the first useful frame, Frame End the last. Don't keep the default values: they know nothing about your shot.
- The right length. Both ends count: from frame 0 to frame 90, that's 91 frames, or 3.8 seconds at 24 frames per second. Each model accepts a minimum and maximum length, and rejects a video outside those limits. On Seedance 2.5 (Reference to Video), a video lasts from 1.8 to 30.2 seconds.
- The right cost. Click Cost Est. again: the estimate takes the chosen part into account.
The setting is in the same place, or nearly, in all your software:
- Resolve, Fusion, Nuke, Flame and Houdini: in the node's panel.
- Blender: in the scene's Start and End range, because the node doesn't have this setting.
The prompt
Describe what you want to see, simply. Some models also have a Negative Prompt, for what you don't want to see.
The fields in your software are often small and not much fun to type in. Edit opens our prompt editor: a real text window, which lets you write much longer prompts, comfortably.
In Resolve, Fusion, Nuke and Flame, this editor also offers Enhance Prompt: AI improves your text for the chosen model. This improvement uses tokens.
The cost
The cost depends on the model and your settings: length, resolution, number of images. After a change, click Cost Est. again. The node then shows an estimate in tokens and your balance.
This figure gives you a reliable guide before you start, but it remains an estimate, not a binding price. We work to keep it as close to reality as possible: it's usually around 90 to 95% of the real price. Once the render is done, we adjust the count to what the model provider actually billed, whether a little more or a little less. You'll find the estimate and the final count for each generation in your personal space.
On Seedance 2.5, Draft creates a 480p preview, for about a fifth of the price of the same video in 1080p. Finalize then turns the preview on screen into a 1080p video, with the same settings.
The result
When the generation is done, the result comes back to your software: in the node, with its versions, and as media under the node. The result and its versions explains everything.