← sources

Need help making my Krea 2 Turbo workflow photorealistic (ComfyUI)

source: reddit (reddit:StableDiffusion)published 2026-11-07 15:53ingested 2026-11-08 02:38

retained text

Hi everyone! I've been working on this workflow for a week, but I'm struggling to get realistic results.

As you can see in the attached output, the image looks washed out and oil-painted, with weird textures around the elevator metal. The jewelry looks like plastic, skin is too smooth, and the lighting and shadows look very artificial.

I'm a beginner with ComfyUI, nodes, models, and LoRAs, so I'm not sure what I'm doing wrong.

I've also been using Claude for help, but I'm going in circles without much improvement.

**I need your help to achieve proper photorealism.**

I'm open to changing my entire workflow, trying different models or LoRAs, or following someone else's workflow.

I've attached my current output and workflow. I'll also share all the models and LoRAs I currently have installed.

Please explain things in simple words, as I'm still learning.

I'd really appreciate some experienced guidance. I want to get this working within the next 1–2 days rather than spending another week guessing.

Here are the Details:

# Krea 2 setup.

# Models

|What|File|Job|
|:-|:-|:-|
|Main model|`selforaKrea2Realistic_v10FP8` (now) or `krea2_turbo_bf16` (before)|Draws the image. Selfora is a realistic version of Krea 2. Turbo is the fast 8-step version.|
|Text reader (CLIP)|`qwen3vl_4b_bf16`, type `krea2`|Turns your prompt into something the model understands.|
|VAE|`qwen_image_vae`|Turns the model's internal picture into a normal image.|

# LoRAs (small add-on files that change the look)

|LoRA|Strength now|Job|
|:-|:-|:-|
|`Famegrid-Natural-V1-Krea-2`|0.25|Natural skin and face look. Too high makes it smooth and painted.|
|`rawsnap v2`|0.55|Phone-camera grain and sharpness.|
|`krea2_identity_edit_v2`|0.25|Keeps faces, straps and the phone shape tidy.|

# Settings now

* Steps 8
* CFG 1.0
* Seed 888 (fixed, so the same input gives the same image)
* Sampler `dpmpp_2m`, scheduler `simple`
* Denoise 1.00
* Size 1.5 MP, 4:5 portrait
* Prompt: the long elevator-selfie text plus the grain words at the end

# Nodes, and why each one is there

|Node|Job|Why we have it|
|:-|:-|:-|
|**Krea 2 + 3 LoRAs (one by one)**|Loads the model and adds the 3 LoRAs, one at a time|I wrote this node. The normal way held two LoRAs in memory at once and froze your Mac (58 GB).|
|**Load CLIP**|Loads the text reader|The model needs it to understand the prompt.|
|**Load VAE**|Loads the VAE|Needed to turn the result into a picture.|
|**Text String (User Prompt)**|Holds your prompt|You type it here.|
|**Boolean (Refine Prompt?)** = off|Would rewrite your prompt|Kept off so your text is sent exactly as you typed it.|
|**CLIP Text Encode (Prompt)**|Turns the prompt into numbers|Required.|
|**Empty Negative Prompt** (`cgi, 3d render, plastic textures…`)|Says what to avoid|Pushes the model away from the plastic and CGI look.|
|**Resolution Selector** and **Latent from Resolution Selector**|Pick the size and make the blank canvas|Sets 4:5 and 1.5 MP.|
|**KSampler**|Does the actual drawing|The main step. It uses the seed, steps and CFG.|
|**VAE Decode (Tiled)**|Turns the result into an image in small tiles|Tiles use less memory.|
|**Save Image**|Saves the picture|Output.|
|**System Monitor**|Shows swap, memory and heat|So you can see when the Mac is under strain.|
|**If/Else Switch, Boolean (Enable LoRA?), Conditioning Zero Out, Generate Text, Concatenate, People Ages**|Leftovers from the original template|Not used in your runs.|

# and here are the LoRA i have downloaded by researching on the reddit that it will help me out.

1 - 2\_vector\_bypass\_lora  
2 - ABOmegle  
3 - DC\_Dark\_ Fantasy\_Style\_4\_\[Krea2Raw\]\[Gpt4.1Captioned\]\[Trigger=dcd@rkf@ntsy.\]\[Rank32\]*v1*  
*4 - DC\_SCI-FI*\[Krea2Raw\]\[Gpt4.1Captioned\]\[Trigger=DCS1F1.\]\[Rank64\]\_v1  
5 - Faded80s  
6 - Famegrid Standard Krea 2  
7 - Famegrid-Natural-V1-Krea-2  
8 - HM1Girl-V1  
9 - Krea2\_P45-NVDStyle\_v1.0  
10 - MW\_analogv1.1  
11 - Playful Eating\_v1.0  
12 - Z-Image-Fun-Lora-Distill-8-Steps\_ComfyUI\_v1  
13 - cinematic\_movie\_still\_krea2  
14 - deedee\_amateur\_photography\_zimage\_base\_and\_turbo\_v1  
15 - famegrid\_spicy\_3167060  
16 - instagram\_aesthetic\_krea2  
17 - krea-detail-enhancer-exp  
18 - krea2RealismV1V2\_krea2RealismV1Lora  
19 - krea2RealismV1V2\_krea2RealismV2Lora  
20 - krea2\_general\_edit\_v2  
21 - krea2\_identity\_edit\_v2  
22 - krea2\_style\_reference  
23 - muscles\_13  
24 - rank64\_turbo\_lora  
25 - rawsnap v2  
26 - ultra\_real\_krea2\_v1  
27 - webcamviewconcept.

document fetched 2026-11-08 00:27 · canonical https://reddit.com/r/StableDiffusion/comments/1xq9ky5/need_help_making_my_krea_2_turbo_workflow