KSampler outputs a denoised LATENT, not a regular pixel image. To save the result as an image in the documented text-to-image workflow, connect that latent to VAE Decode, provide the matching VAE, then connect the decoded IMAGE to Save Image. The nodes handle different representations; they are not interchangeable previews or save buttons.
Read the three types in the workflow
LATENT is the sampler's result in latent space: the representation the diffusion process denoises. IMAGE is the pixel-space image representation used for viewing and saving in this workflow. VAE is the model component used to convert between those spaces. Accordingly, the VAE Decode node needs both a latent input and a VAE input; its output is an image.
The official ComfyUI text-to-image guide describes KSampler as producing a latent, VAE Decode as converting that latent to pixel space, and Save Image as saving the decoded result. That is why a completed KSampler alone does not mean the workflow has produced the ordinary image you intend to save.
Wire the decode and save path
Use this ordered check in the workflow editor:
- Find the
LATENToutput onKSampler. It is the denoised result you want to convert. - Connect that output to the latent input on
VAE Decode. Do not substitute a conditioning or model connection. - Connect a compatible
VAEoutput, commonly from theLoad Checkpointnode in the documented example, to theVAE Decodenode's VAE input. The guide identifies the checkpoint as a source of a VAE, but this article does not prescribe a separate VAE-selection rule for every model or workflow. - Connect the
IMAGEoutput fromVAE Decodeto the image input onSave Image. - Queue the workflow. When processing completes, inspect the
Save Imagenode's preview; the guide says the node saves the decoded image toComfyUI/output.
Check the connections by type
The table uses hypothetical port observations. It is a wiring check, not a report of a software test. Read the connection at each endpoint, not only the visual left-to-right order of the nodes.
| Checkpoint | Hypothetical observation | Expected connection or next action |
|---|---|---|
| KSampler result | Output is labeled LATENT | Connect it to VAE Decode's latent input |
| Decoder input | Latent input is connected; VAE input is empty | Add a VAE connection to the VAE input |
| Decoder result | Output is labeled IMAGE | Connect it to Save Image's image input |
| Save node | Image input is empty | The decoded image cannot reach this save node; connect the decoder output |
| Completed run | Save Image shows a preview | This is the documented observable checkpoint for a decoded image in this workflow |
If the image does not appear
If KSampler finishes but Save Image has no preview, follow the data path in order: confirm the sampler's LATENT is connected to the decoder's latent input; confirm the decoder also has a VAE; then confirm the decoder's IMAGE reaches the save node's image input. A missing connection at any of those points means the documented path is incomplete.
If all three connections are present but the run still does not show a saved image, the source guide does not provide enough information to diagnose every model, custom node, or runtime failure. Check the actual node's reported error and the workflow's model and VAE setup. Do not infer from the guide that changing sampler settings will repair a missing decode or save connection.
Common questions
Can I connect KSampler directly to Save Image?
Not in the documented path described here: KSampler supplies LATENT, while the save node receives the decoded IMAGE. Put VAE Decode between them and provide its VAE input.
Does VAE Decode generate a new image independently?
Its documented role is to convert the sampler's latent into pixel space. It is a representation-conversion step, not a replacement for the denoising that KSampler performed.
Where should I look for the saved result?
The guide says Save Image saves the decoded image in the local ComfyUI/output folder and presents a preview in the node interface after the workflow completes. The exact behavior of other workflows or custom save nodes is outside that guide's stated scope.
Sources and next step
This guide interprets public documentation and uses explicitly hypothetical acceptance examples, not application tests. Controls and availability can change; verify the cited manual against your installed version.
For work created in Imagild, carry the input requirements and acceptance checklist into the photo editor. The external software controls discussed above belong to their respective products.