Following my few videos about Magnific.ai, many of you commented that the price was too high and wanted a free alternative to Magnific.ai Looks like we have it and its by a @Fictions-ai (Thibaud Zamora) who released a ComfyUI Upscaler Workflow that hallucinates and creates highly detailed upscaled images using your own instance of ComfyUI and your own hardware (if you have it) or you use cloud GPU.
ComfyUI Workflow: https://github.com/fictions-ai/sharing-is-caring
TurboVisionXL SDXL Model: https://civitai.com/models/215418/turbovisionxl-super-fast-xl-based-on-new-sdxl-turbo-3-5-step-quality-output-at-high-resolutions
Add Detail XL LoRA: https://huggingface.co/PvDeep/Add-Detail-XL/tree/main
4x NMKD Upscaler: https://civitai.com/models/141491?modelVersionId=156841
The workflow is easy to understand and most of the settings are automatically computed. All you need to do is adjust the Hallucination amount which is recommended from 0.15 – 0.40, in some of my tests I even reduced it down to 0.10.
My initial thought was to re-run the same images that I had run during my comparison with Midjourney and Magnific AI in my last post so I can get a feel for how good this workflow is.
So far, I think the upscale is better matched to your original image if you use the prompt or main subject prompt from your original image. Without the prompt the upscale may drift away from the original subject a bit or a lot depending upon the Hallucination level.
It serves as genuine alternative to Magnific AI which was deemed to expensive by many of the readers and viewers.
Comparison Results
Below images are comparison between the original image produced in Midjourney and then upscaled using this workflow in ComfyUI. The images were reduced in size to render for the site.
One of the differences I can see is the images lose some of their vibrance and appear duller than the original. I think this can be adjusted in post processing (Colour Match in Photoshop) or Lightroom editing. However I think if I play with the Steps and/or CFG Scale setting of the Ultimate SD Upscaler node it might increase the contrast and vibrance details.
The woman’s face has changed a bit in the upscaled image but there is resembles to the original
The pirate looks pretty detailed, the face and eyes have lots of definition along with the hair and clothes. The model seems to have brought realistic details in this upscaled image.
Overall the details and quality of the upscale does serve as wonderful alternative to Magnific AI. It is not as refined as the results from Magnific AI team but I’m sure Thibaud or other community members can take this workflow further.
I will continue experimenting more and sharing more results on the blog and socials.
If you'd like to support our site please consider buying us a Ko-fi, grab a product or subscribe. Need a faster GPU, get access to fastest GPUs for less than $1 per hour with RunPod.io
What does hallucination node does?
I add more or less new details in the image. The node is just a value setting which is passed to Ultimate SD Upscaler, the node itself doesn’t do much other than specify a decimal value.
I got this error when I drag and drop the image workflow. “Custom validation failed for node: image – Invalid image file: ComfyUI_temp_kqydq_00005_b.png”. Do you know how to fix it?
You should contact the creator via the GIT page and report a bug. I am afraid I cannot help with this other than suggest you update ComfyUI and check if there are any errors in the workflow that you imported
when i use your workflow i have this problem message:
Prompt outputs failed validation
Text Concatenate:
– Value not in list: clean_whitespace: ” not in [‘true’, ‘false’]
Text Concatenate:
– Value not in list: clean_whitespace: ” not in [‘true’, ‘false’]
CheckpointLoaderSimple:
– Value not in list: ckpt_name: ‘turbovisionxlSuperFastXLBasedOnNew_tvxlV32Bakedvae.safetensors’ not in [‘OfficialStableDiffusion\\sd_xl_base_1.0.safetensors’, ‘SDXL-TURBO\\sd_xl_turbo_1.0_fp16.safetensors’]
UpscaleModelLoader:
– Value not in list: model_name: ‘4x_NMKD-Superscale-SP_178000_G.pth’ not in [‘4xNMKDSuperscale_4xNMKDSuperscale.pth’, ‘8x_NMKD-Superscale_150000_G.pth’]
Have you downloaded and placed those models in the respective folders? If so then click on Refresh and then select the model again. It can happen that the model its trying to use it not there or not in the same directory or same filename.