For a large reference image, consider placing ComfyUI’s FluxKontextImageScale node before the image enters the FLUX Kontext editing path when you want its input dimensions adjusted to a documented training-size preset. The node chooses a preset based on aspect ratio and uses Lanczos scaling. Check the resulting dimensions and composition: a ratio-based choice is not a guarantee that the original framing or fine detail is preserved. The documentation does not specify how it resolves ties or rounds between presets, so inspect the actual output rather than predicting a particular selection.
Decide whether to normalize, then inspect the result
The node is worth considering when the reference is unusually large, or when a direct edit produces unwanted behavior such as multiple subjects appearing. The official documentation identifies oversized inputs as a possible quality issue, not a certainty. If the image is already a manageable size and the composition is important, first compare its ratio with the listed presets; do not add resizing just because the input is large in file size.
Use this sequence:
- Record the reference width and height, and calculate width divided by height. Note the subject’s position and any edges or details that must remain visible.
- Compare that ratio with the documented preset list. Decide whether testing the dedicated node is warranted, especially for an oversized input.
- If you test the node in your actual workflow, check that the reference image is connected to
FluxKontextImageScale'sIMAGEinput and that its resizedIMAGEoutput is connected to the subsequent edit path. The documentation identifies these input and output data types but does not specify a universal workflow order or the appearance of connection labels. - Read the output image’s actual width and height. Record its ratio too; do not infer the selected preset from the input ratio.
- Compare source and output framing at the same view scale. Check subject edges, spacing, and small details, then decide whether the resized reference still serves the edit.
This node is a reference-input size adjustment. It is not documented as a total-pixel-budget setting or as a longest-edge limiter. For its scaling method and ratio-based selection, see the official node documentation.
Hypothetical ratio-to-preset worksheet
The following is a hypothetical acceptance exercise, not a node run or a guarantee of which dimensions the node will choose. A 2400 × 1600 source has a ratio of 1.500. The list includes 1.500 and the adjacent wider preset 1.660. The documentation says the closest ratio is selected, but does not define tie-breaking or the selection calculation in enough detail to promise an outcome for every input. Record what the node actually outputs.
| Check | Hypothetical observation | Decision |
|---|---|---|
| Source | 2400 × 1600; ratio 1.500 | Keep this as the comparison baseline |
| Nearby presets | 1.500 and 1.660 appear in the documented list | Treat them as candidates, not a predicted selection |
| Node output | Suppose inspection reports 1248 × 832; verify in the workflow | Compare measured dimensions and ratio with the source |
| Composition | Suppose the subject’s left edge remains visible, but fine texture is less distinct | Accept only if framing and required detail still suit the edit |
| Task requirement | Suppose the edit depends on a small label in the original | Do not assume resizing retains readable source detail; reassess the input or workflow |
The observations in the last three rows are invented for the worksheet. They describe what to check, not expected behavior. Keep separate notes for dimensions, visible framing, and the detail the edit requires.
Limits and what to do when it fails
If the output’s framing or retained detail does not suit the edit, compare it with the source and reconsider whether this node belongs on that reference path. Resizing cannot restore content missing from the source.
Record composition and prompt adherence as separate pass/fail checks: a prompt miss does not establish a sizing problem, and prompt adherence does not make unsuitable framing acceptable. The documentation specifies the node’s IMAGE input and output, Lanczos scaling, and ratio-based preset selection; it does not specify prompt controls or tie-breaking behavior.
Three quick questions
Is this the same as resizing by longest edge? No documented rule here caps the longest edge. The node selects among listed training-size presets based on input aspect ratio.
Does resizing guarantee that all source detail is preserved? No. Compare the output with the source at the detail level your edit requires; do not assume resizing retains small or fine details.
Should I change the prompt if the subject looks distorted? First compare source and node output dimensions and framing. If that comparison passes, review prompt adherence separately; dimensions and instruction-following are different checks.
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.