YES SUCCEEDED!!!

Just rendered an image at 944×1152 (slightly above 1024×1024) using Flux1-Schnell-FP8 on my 6700 XT, and it works! (Image 1 is the Real-ESRGAN 2× upscaled version)

Workflow 1: Sampling (Image 2)

Prompt executed → UNet generates the latent

Step 1 (model load + latent generation) took 419 seconds

Output: Latent tensor saved to disk

Workflow 2 : VAE Decode (Image 3)

Latent loaded → VAE decodes the image

Duration: 7.5 seconds

Advantage: UNet doesn’t need to stay in VRAM → VRAM freed, even on 12 GB GPUs

The problem with the stock LoadLatent Node

Dropdown only shows files if they were produced / annotated by a previous SaveLatent Node

Node is designed to pass latents inside a graph, not load arbitrary files from disk

Purpose: prevents accidentally loading wrong files

Workaround (Image 4)

Edited /ComfyUI/nodes.py, class LoadLatent

Hardcoded latent path → Node now loads directly from disk

Result: Workflow 2 runs instantly, UNet can be unloaded

Timing

Step 1 (model load + latent generation): 419 s

Step 2 (VAE decode): 7.5 s

Result: High-res images on a 12 GB RDNA2 GPU are now possible on Flux1-Schnell-FP8 without ComfyUI crashing! (Image 5 is the original output)

This might actually become my new Flux workflow: render quick 512×512 previews first (which works perfectly on RDNA2 GPUs), sort out the good ones, extract the seed from the PNG metadata, and then re-render only the selected images with the same seed using the split workflow at higher resolutions. This way, high-resolution Flux1-Schnell-FP8 renders become possible on 12 GB RDNA2 GPUs D:

Question at the end: Has anyone ever done this before? Because I have no clue xD

#ComfyUI #flux #Flux1SchnellFP8 #FP8 #AMD #RDNA2 #VAE #AIArt #Pixelfed #HighResolution #GPUOptimization #LatentWorkflow #AIWorkflow #AIHacks #RealESRGAN #Upscale #AIExperiment #CreativeAI #DigitalArt #AICommunity #python #linux #opensource #foss

One year ago today I opened my Pixelfed profile 🎉

Time for a short retrospective of how it all began.

Late 2024, 2 a.m.: I was manually integrating peaks from chromatograms in Chromeleon when I thought: why can’t an AI do this?
The idea didn’t go anywhere, but I started exploring AI frameworks and ended up with image generation. ROCm on Debian, EasyDiffusion, and then Pixelfed.

Later Debian and ROCm drifted apart, so I posted some real-life photos. With an Ubuntu chroot, everything ran cleanly again, even AUTOMATIC1111. SD 1.5 was my standard for a long time. Early this year I tried FLUX in ComfyUI but had to drop it: RDNA2 + no FP8 + incomplete HIP → FLUX-VAE not practical. Mid-January I finally fixed the NaNs in SDXL VAE in A1111.

Now I’m fully on ComfyUI, can render 1024×1024, and 512+ px no longer OOMs.

End of 2025, I used Pixelfed for IT/FOSS networking: the FOSS Advent Calendar. Posts were seen thanks to ActivityPub, and I even started my own dev blog xD

Thanks 💜 to everyone who follows me, especially my regular viewers and those I really exchange with.
Pixelfed remains my place to share, experiment, and learn.

1 year on Pixelfed, and it all started with peaks at 2 a.m.

tl;dr: Thanks so much to everyone who follows me, especially my regular viewers and those I really exchange with, you are awesome (in Austrian slang: Ihr seid ur leiwand 💜)

#Pixelfed #Fediverse #OpenSource #FOSS #Anniversary #1Year #Celebration #Birthday #Milestone #BirthdayCake #Fireworks #Festive #Colorful #AI #AIArt #GenerativeArt #ComfyUI #SDXL #StableDiffusion #ROCm #Linux #ThankYou #AiCommunity

Finally got SDXL running!

I looked into the error
"modules.devices.NansException: A tensor with NaNs was produced in VAE"
and here’s what it means:

Briefly, how an image is generated with a diffusion model: The text encoder interprets the prompt, the UNet "dreams" iteratively in the latent space from noise into an image structure, and the VAE translates this latent vision into visible pixels.

A tensor is simply a multi-dimensional array of numbers, basically the data structure where the model stores all its calculations, like colors, intensities, and intermediate results of the image.

In this case, the VAE experienced a numerical instability: the latent tensor contained invalid values (NaNs), so the dreamed image could not be decoded correctly. In short: the model was still dreaming in the latent space, but the numbers “exploded” along the way (e.g., division by zero, overflow, or undefined operations).

#StableDiffusion #SDXL #AIArt #DiffusionModel #VAE #LatentSpace #Tensor #DigitalArt #CinematicArt #Kunst #KI #AI #DigitalIllustration #StilizedRealism #UrbanFantasy #Motion #DramaticLighting #FilmStill #AICommunity

🚀 Tired of AI hype? Emanon drops a weekly, no‑fluff newsletter + podcast that actually helps builders implement AI—plus 1‑hr consulting, tools benchmarks, and a global Discord community. Join the signal‑only zone. #AI #NoHype #Emanon #Tech #AICommunity