A Krea 2 text-to-image workflow for ComfyUI, the one I have been playing with. One prompt box, a LoRA stack, and four switches that decide how much work happens after the base render. One PNG per run, with the generation data and the workflow inside.
I benchmarked every switch: what each one costs and what it changes is in <a target="_blank" rel="ugc" href="https://civarchive.com/articles/38721">the comparison article</a>. The short version is at the bottom of this page.
<h3>The four switches</h3>
<ul>
<li>Use LoRAs: on, the model and the text encoder go through the LoRA stack, and the LoRA names go into the saved prompt for the metadata. Off, both skip the stack.</li>
<li>Use bf16 Text Encoder: on, the bf16 Qwen3-VL 4B encoder. Off, the int8 one, about half the size. Take int8 when VRAM is tight.</li>
<li>Use Hires Fix: on, the base picture goes through 4x-UltraSharp to 1.5x, then 6 more steps at denoise 0.35, with tiled VAE encode and decode. 1200x1600 comes out 1800x2400.</li>
<li>Use Face Detailer: on, FaceDetailer finds faces with face_yolov8m and redraws them, 8 steps at denoise 0.35. It works on the hires picture when the hires fix is on, and on the base picture when it is off.</li>
</ul>
In the file all four are on, so you get the full stack.
<h3>Defaults</h3>
<ul>
<li>1200x1600 from CR Aspect Ratio (custom size). 10 steps, CFG 1, <code>er_sde</code>, <code>simple</code>.</li>
<li>No negative prompt box. At CFG 1 the negative has no effect, so the graph feeds the sampler an empty one.</li>
<li>One Seed node feeds the base pass, the hires pass and the face detailer.</li>
<li>The checkpoint loads through Image Saver's UNet loader, so the model name lands in the metadata.</li>
<li>Image Saver writes the prompt, LoRA names, seed, steps, CFG, sampler and scheduler into the PNG and embeds the workflow. The file name is time, model name and seed.</li>
<li>A notes node holds the trigger words of my three character LoRAs (<a target="_blank" rel="ugc" href="https://civarchive.com/models/3005329">Shawy</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/3007185">Zara</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/3008752">Aria</a>) and my house style sentence. It is wired to nothing. Copy from it.</li>
</ul>
<h3>LoRA slots</h3>
The stack ships with my <a target="_blank" rel="ugc" href="https://civarchive.com/models/3005329">Shawy LoRA</a> at 0.7 in slot 1, plus two style LoRAs from my stack at 0.4 and 0.3. Swap in your own. My files sit in a subfolder, so open each slot once and pick your file.
Gotcha: set any LoRA you do not have to <code>None</code>. ComfyUI checks every dropdown before it runs, so a missing file stops the queue even with Use LoRAs off.
<h3>Required custom nodes</h3>
<ul>
<li><a target="_blank" rel="ugc" href="https://github.com/Suzie1/ComfyUI_Comfyroll_CustomNodes">Comfyroll Studio</a>: Image Size, the LoRA stack, Apply LoRAs, the metadata prompt join, the 1.5x upscale.</li>
<li><a target="_blank" rel="ugc" href="https://github.com/alexopus/ComfyUI-Image-Saver">ComfyUI Image Saver</a>, 1.23.0 or newer: the model loader, steps, CFG, sampler, scheduler, metadata and Save Image.</li>
<li><a target="_blank" rel="ugc" href="https://github.com/giriss/comfy-image-saver">comfy-image-saver</a> by giriss: the Seed node only. A different pack from the one above, with almost the same name.</li>
<li><a target="_blank" rel="ugc" href="https://github.com/ltdrdata/ComfyUI-Impact-Pack">Impact Pack</a>: Face Detailer.</li>
<li><a target="_blank" rel="ugc" href="https://github.com/ltdrdata/ComfyUI-Impact-Subpack">Impact Subpack</a>: the face detector.</li>
<li><a target="_blank" rel="ugc" href="https://github.com/ltdrdata/was-node-suite-comfyui">WAS Node Suite</a>, the maintained copy: Text Multiline, the prompt box.</li>
<li><a target="_blank" rel="ugc" href="https://github.com/jags111/efficiency-nodes-comfyui">Efficiency Nodes</a>: LoRA Stack to String, the LoRA names for the metadata.</li>
</ul>
The switches, the booleans and the notes node are core ComfyUI. If they show red, update ComfyUI. Tested on ComfyUI 0.44.0.
<h3>Model files</h3>
<ul>
<li>Checkpoint: <a target="_blank" rel="ugc" href="https://civarchive.com/models/958009?modelVersionId=3139241">RedCraft 3.0 (Krea 2)</a>, the int8 build, <code>redcraft23_30Krea2_int8_convrot.safetensors</code>, in <code>models/diffusion_models</code>.</li>
<li>Text encoder: <code>qwen3vl_4b_bf16.safetensors</code> or <code>qwen3vl_4b_int8_convrot.safetensors</code>, in <code>models/text_encoders</code>, loader type <code>krea2</code>. Both loaders must point at a file that exists. With one encoder only, pick it in both.</li>
<li>VAE: <code>qwen_image_vae.safetensors</code>, in <code>models/vae</code>.</li>
<li>For the hires fix: <code>4x-UltraSharp.pth</code> from <a target="_blank" rel="ugc" href="https://openmodeldb.info/models/4x-UltraSharp">OpenModelDB</a>, in <code>models/upscale_models</code>.</li>
<li>For the face detailer: <code>face_yolov8m.pt</code> from <a target="_blank" rel="ugc" href="https://huggingface.co/Bingsu/adetailer">Bingsu/adetailer</a>, in <code>models/ultralytics/bbox</code>.</li>
</ul>
<h3>What the benchmark says</h3>
From <a target="_blank" rel="ugc" href="https://civarchive.com/articles/38721">the comparison article</a>, 89 renders on one RX 7900 XTX:
<ul>
<li>At feed size the base picture is hard to tell from the full stack. On my card the base took about 36 s, everything on about 177 s.</li>
<li>The hires fix pays when a picture is viewed big or has small details. In the test it turned a garbled pendant into a clean amber stone.</li>
<li>To save about 70 s per hires render, swap the 4x-UltraSharp node for core <code>Upscale Image By</code>, set to <code>lanczos</code> and <code>scale_by</code> 1.5. At viewing size it looked the same, a little softer on photo hair. Two cases, one seed each.</li>
<li>The face detailer at 0.35 changed my LoRA characters' faces: older on the photo look, red eyes toward brown, mottled cel shading. For LoRA characters, leave it off. Lower denoise I did not test.</li>
</ul>
<h3>Known limits</h3>
<ul>
<li>Tested on one AMD card under ROCm. Nothing in the graph is AMD-specific.</li>
<li>Tested on RedCraft 3.0 only, with two character LoRAs in three styles.</li>
</ul>
You are free to try it. Tell me in the comments if it works for you. If it breaks, paste the error text.

Krea 2 All-in-One: LoRA Stack, Hires Fix, Face Detailer

v1.0. The workflow the <a target="_blank" rel="ugc" href="https://civitai.red/articles/38721">comparison article</a> tests, with four switches: Use LoRAs, Use bf16 Text Encoder, Use Hires Fix, Use Face Detailer. All four are on in the file.
Base 1200x1600, 10 steps, CFG 1, <code>er_sde</code>, <code>simple</code>. Hires fix: 4x-UltraSharp to 1.5x, then 6 steps at denoise 0.35, final 1800x2400. Face Detailer with face_yolov8m, 8 steps at denoise 0.35. LoRA slot 1 is my <a target="_blank" rel="ugc" href="https://civitai.red/models/3005329">Shawy LoRA</a> at 0.7, slots 2 and 3 two style LoRAs at 0.4 and 0.3.
Tested on RedCraft 3.0 (int8) in ComfyUI 0.44.0.

Files with SHA256: 4ad62058a8b1f85052977ce071eca93981c4b7b1655748f497c6623cb5e498f9