There are three parts. The first step is to use flux to output the image in 8 steps. The second step is to use 2x magnification + block magnification and flux block repair. The third step is to add the icing on the cake and add details in latent. I found that traditional flux magnification does not really increase the pixels and is very stuck. According to my research, block magnification is the best way to bring out the power of flux, so I made this workflow.
FLUX-super upscale 超级放大
Workflow Preview
Showcases (Image/Video)

Showcases (Image/Video)

Nodes Detail
68 NodesPrimitive Node Types
20- PreviewImage1
- ConditioningConcat1
- VAEEncode3
- VAELoader1
- VAEDecodeTiled2
- SaveImage3
- VAEDecode1
- CLIPTextEncode3
- ImageUpscaleWithModel1
- LatentUpscaleBy1
- LoadImage1
- LoraLoader1
- UpscaleModelLoader1
Custom Node Types
48- RandomNoise2
- Display Any (rgthree)1
- easy imageBatchToImageList1
- TTP_CoordinateSplitter1
- ImageSmartSharpen+1
- Reroute6
- TTP_Image_Tile_Batch1
- KSamplerSelect2
- InjectLatentNoise+1
- easy imageListToImageBatch1
- CR Prompt Text1
- SamplerCustomAdvanced2
- FluxGuidance2
- BasicGuider2
- BasicScheduler2
- Text List1
- Text List to Text1
- Florence2Run2
- Florence2ModelLoader2
- ShowText|pysssss2
- TiledDiffusion1
- Anything Everywhere1
- Image Comparer (rgthree)3
- TTP_condsetarea_merge1
- TTP_condtobatch1
- KSamplerAdvanced //Inspire1
- JWImageResizeByLongerSide1
- ImageScaleToTotalPixels1
- TTP_Image_Assy1
- UNETLoader1
- DualCLIPLoader1
- TTP_Tile_image_size1