toobusy Hires Upscale:
ToobusyHiresUpscale prepares an image for a higher-resolution refinement workflow. It enlarges the image with an upscale model, optionally resizes that result, and encodes the finished image into a latent representation for further image-generation work. By combining these steps in one node, it reduces the number of connections you need while letting you choose between the model’s full enlargement and a more controlled final size.
toobusy Hires Upscale Input Parameters:
image
The IMAGE to enlarge. Its starting dimensions, together with the upscale model and any subsequent resizing, determine the final image size. Connect the image you want to prepare for higher-resolution refinement; this input has no default value.
vae
The VAE used to encode the final resized image into a LATENT. Use the VAE appropriate for the downstream workflow that will consume that latent. This required input has no default value.
upscale_model_name
The name of the upscale model to load when no upscale_model connection is provided. The available options come from the installed upscale_models list; if that list cannot be read or is empty, the interface falls back to ESRGAN/4x_foolhardy_Remacri.pth. The default is chosen from preferred available models, then an available name containing “remacri,” or otherwise the first available name. This setting is ignored when upscale_model is connected. It has no numeric minimum or maximum.
downscale_method
The resampling method used if the model-upscaled image is resized afterward. Options come from the available ImageScaleBy methods; if they cannot be read, the interface offers nearest-exact, bilinear, area, bicubic, and lanczos. The default is lanczos when available, or the first available method otherwise. This choice does not affect the upscale-model step and is not used when scale_by is 1.0.
scale_by
The factor applied to the image after model upscaling. Its default is 0.50, its minimum is 0.05, its maximum is 8.0, and its adjustment step is 0.05. For example, after a 4× model upscale, 0.50 targets approximately 2× the starting dimensions; 1.0 keeps the model’s output without a second resize. Values below 1.0 reduce the model-upscaled image, while values above 1.0 enlarge it further. The resize step is skipped when the value is effectively 1.0.
upscale_model
An optional UPSCALE_MODEL connection that overrides upscale_model_name. Connect a model you have already loaded when you want to use that specific model rather than have this node load one by name. When this connection is absent, the node loads the named model instead. There is no default connected model.
toobusy Hires Upscale Output Parameters:
image
The final IMAGE after model upscaling and, unless scale_by is 1.0, resampling. This is the image that the node passes to VAE encoding. Use it when you need to view, save, or route the prepared image elsewhere in your workflow.
latent
The LATENT produced by encoding the final image with the supplied VAE. Connect it to a compatible downstream refinement or sampling workflow when you want to continue working from the enlarged image in latent form.
width
The final image width in pixels, returned as an INT. It is read from the image after all scaling, so it describes the image this node returns rather than the original input. If the node cannot read the image dimensions, it returns 0 for both width and height.
height
The final image height in pixels, returned as an INT. Like width, it reflects the image after all scaling and can help you set or check dimensions elsewhere in your workflow. If the node cannot read the image dimensions, it returns 0 for both width and height.
toobusy Hires Upscale Usage Tips:
- For a roughly 2× final enlargement with a 4× upscale model, start with
scale_byset to 0.50. Check thewidthandheightoutputs for the actual resulting dimensions. - Set
scale_byto 1.0 when you want the model’s full output and do not need the additional resampling step. - Choose
downscale_methodbased on how you want the post-model resize to look. It only matters whenscale_byis not 1.0. - If an upscale model is already loaded elsewhere, connect it to
upscale_model. The connected model takes priority, so you do not need to changeupscale_model_nameto select it. - Keep the final image size in mind when increasing
scale_by: larger images can require more memory and processing time during upscaling, resizing, and VAE encoding.
toobusy Hires Upscale Common Errors and Solutions:
Upscale model not found or failed to load
- Explanation: When
upscale_modelis not connected, the node asksUpscaleModelLoaderto loadupscale_model_name. An unavailable or unreadable model can prevent processing. The exact error text depends on the underlying loader. - Solution: Select a model that is actually installed in your upscale-model list, or connect a working
upscale_modeldirectly. If you recently added a model, refresh the available model list before selecting it.
Out of memory during upscaling or VAE encoding
- Explanation: A large input image, a high enlargement factor, or a
scale_byvalue above 1.0 can produce an image that exceeds available memory. The exact message depends on the component that runs out of memory. - Solution: Try a smaller starting image or lower
scale_byvalue. Avoid further enlargement above 1.0 unless you need it and have enough memory.
VAE encoding failed
- Explanation: The node encodes the final image using the connected
vae. If that connection is missing or the underlying VAE encoder cannot process the image, the node cannot produce its latent output. The exact error text comes from the encoder or workflow. - Solution: Check that
vaeis connected to a working VAE appropriate for your downstream workflow, then try again with a smaller image if memory may be the issue.
Invalid or unavailable resampling method
- Explanation: When
scale_byis not 1.0, the node sendsdownscale_methodtoImageScaleBy. A method that is not supported by the current installation can cause that resize step to fail. - Solution: Choose a method currently offered by the
downscale_methodmenu, or setscale_byto 1.0 if you do not need post-model resizing.
width and height are 0
- Explanation: The node returns 0 for both dimensions if it cannot read the final image’s shape. This is a fallback value, not a measured image size.
- Solution: Check that the image-producing steps return a valid IMAGE and that the connected model is working. Inspect the
imageoutput before relying on the dimension outputs.
