← Back to Guides
Krea 2 detail-boost workflow guide: Enhancer node, 2× combo pass, and IdeoKrea LoRA
This guide turns Aitrepreneur’s Krea 2 update video into a practical, safety-aware ComfyUI workflow playbook. It focuses on three upgrades: replacing the older Krea 2 rebalance approach with a lower-strength Krea2T Enhancer, using a two-pass “2× combo” workflow to sharpen and add texture, and applying the experimental IdeoKrea LoRA for a higher-contrast Ideogram-like style.
Source: AitrepreneurVideo date: 2026-06-27Runtime: 11:10Difficulty: intermediateLocal ComfyUI workflow
Safety boundary: the source video discusses adult-capable/local image generation and filter-bypass tools. This guide preserves the technical workflow concepts but avoids explicit sexual prompt recipes. Use these tools only for legal, consensual, age-appropriate, non-abusive work. Do not create sexualized minors, non-consensual intimate imagery, impersonation abuse, harassment, or other illegal/harmful content.
Quick outcome
You will build:
A Krea 2 ComfyUI workflow that can generate a first image, optionally strengthen prompt following, and then run a light second Krea 2 Turbo refinement pass.
Best use case:
When you already like Krea 2’s composition and aesthetics but want more texture, sharper details, stronger prompt adherence, or an Ideogram-like contrast/style treatment.
Core rule:
Keep a clean baseline workflow, then add one enhancement at a time: Enhancer, combo pass, or LoRA. Compare with a fixed seed before deciding what actually improved.
Table of contents
- What changed from the previous Krea 2 workflow
- Prerequisites and file placement
- Build a stable baseline first
- Use the Krea2T Enhancer for prompt following
- Set up the 2× combo second-pass workflow
- Use the IdeoKrea LoRA
- Recommended comparison settings
- Troubleshooting
- Verification checklist
- Sources and useful links
1. What changed from the previous Krea 2 workflow
The video is an update to Aitrepreneur’s earlier Krea 2 guide. The main claim is not that Krea 2 changed as a base model, but that the surrounding workflow can now produce better images with small changes.
| Change | Old/previous approach | Updated approach in this guide | Why it matters |
|---|
| Prompt-following / filter pressure | Conditioning Krea 2 Rebalance style node. | ComfyUI Krea2T Enhancer / Krea2T Enhancer-style node at low strengths such as 0.1. | The presenter reports better preservation of the original realistic style while nudging the model to follow difficult prompt details. |
| Detail boost | Single Krea 2 text-to-image pass. | A “2× combo” flow: generate the first image, then refine that same image with a light Krea 2 Turbo pass using the Turbo LoRA setup. | The second pass can add sharper textures and extra detail while keeping composition close to the first pass. |
| Style option | Krea 2 alone or community LoRAs. | Optional IdeoKrea LoRA, trained to push Krea 2 toward an Ideogram 4-like visual treatment. | Useful when you want more contrast, color pop, poster-like sharpness, and sometimes better text/graphic feel. |
| Community ecosystem | Few LoRAs available immediately after release. | More community LoRAs on Civitai and related sites. | Greater style and subject coverage, but also more need to verify compatibility and provenance. |
2. Prerequisites and file placement
This update assumes you already have a working Krea 2 ComfyUI installation. If not, start with the earlier setup guide/video and the official model page before attempting the combo workflow.
- ComfyUI updated. Update ComfyUI before loading newer workflow files or custom nodes.
- Krea 2 model files installed. The official ComfyUI model hub is Comfy-Org/Krea-2 on Hugging Face.
- Qwen text encoder and Qwen image VAE available. Krea 2 workflows typically depend on Qwen-family prompt encoding and the Qwen image VAE.
- Krea 2 Turbo model path working. The combo workflow relies on a Krea 2 Turbo generation/refinement path rather than only the raw/base model.
- LoRA folder ready. Put Krea 2-compatible LoRAs in
ComfyUI/models/loras/, then refresh or restart ComfyUI.
- Custom node folder ready. Put custom nodes under
ComfyUI/custom_nodes/, install their requirements if provided, then restart ComfyUI.
Useful file-placement map
| Asset | Purpose | Typical location | Notes |
|---|
| Krea 2 / Krea 2 Turbo model | Primary diffusion/image model. | ComfyUI/models/diffusion_models/ | Use the model variant your workflow expects. Turbo is the practical starting point for fast generation and the combo refinement path. |
| Qwen text encoder | Prompt/text conditioning. | ComfyUI/models/text_encoders/ | If conditioning shape errors occur, the selected encoder may not match the workflow/model. |
| Qwen image VAE | Decode latents to images. | ComfyUI/models/vae/ | Missing/wrong VAE usually causes red-node load errors or failed decoding. |
| Krea2T Enhancer custom node | Optional prompt-following/conditioning adjustment. | ComfyUI/custom_nodes/ | Restart after install. Use low strength first. |
| Turbo LoRA / Krea 2 compatible LoRAs | Style/control adapters and combo support. | ComfyUI/models/loras/ | Check each LoRA’s docs for trigger phrases and recommended strength. |
| IdeoKrea LoRA | Ideogram-like style treatment for Krea 2. | ComfyUI/models/loras/ | The video describes it as a test/experimental version; compare before and after before relying on it. |
3. Build a stable baseline first
Before adding the new pieces, confirm that normal Krea 2 text-to-image generation works. This gives you a control image for meaningful comparison.
A. Load your Krea 2 workflow
- Start ComfyUI.
- Load your known-working Krea 2 text-to-image workflow or the updated workflow from Aitrepreneur’s install guide if you have access to it.
- Select the Krea 2 model, text encoder, and VAE files explicitly.
- Disable all optional LoRAs and enhancer/rebalance nodes for the first test.
B. Run one safe baseline prompt
Use a prompt that tests detail and style but does not rely on edge-case content. Example:
a cinematic close-up photo of an ornate brass pocket watch on a dark wooden desk, fine engraved details, soft window light, shallow depth of field, realistic textures
Save the prompt, seed, width/height, steps, CFG, sampler, scheduler, and model names. You will reuse them for comparison.
Why this matters: if you change the seed, prompt, LoRA, enhancer, and sampler all at once, you cannot tell which piece improved or damaged the image. Use fixed comparisons.
4. Use the Krea2T Enhancer for prompt following
The video replaces the older Krea 2 rebalance-style node with a newer Krea2T Enhancer-style node. The purpose is to make Krea 2 follow difficult prompt details more reliably while preserving the original style better than the older rebalance method.
Recommended starting method
- Duplicate your baseline workflow and rename it clearly, for example
krea2-enhancer-test.json.
- Add the Krea2T Enhancer node where the updated workflow expects it, generally in the positive-conditioning path before sampling.
- Set the enhancer strength low first. The presenter demonstrates
0.1 as enough to change difficult prompt adherence while retaining realism.
- Run the exact same prompt and seed as your baseline.
- Compare composition, realism, texture, color, and whether the specific hard-to-follow detail appeared.
| Strength range | What to expect | When to use |
|---|
0.05–0.15 | Subtle nudge; often preserves the baseline look best. | Start here for prompt-following fixes or detail compliance. |
0.2–0.5 | More visible change to composition/style. | Use when low strength fails and you accept stronger image drift. |
0.6–1.0 | High intervention; may noticeably alter realism or style. | Advanced testing only; compare carefully against baseline. |
Quality tradeoff: increasing enhancer strength may improve adherence to a specific phrase, but it can also change the image’s look. The video’s main point is that low values can be useful without the older “plasticky” quality shift.
5. Set up the 2× combo second-pass workflow
The “2× combo” is the headline update. It generates a normal Krea 2 image first, then sends that generated image through a second Krea 2 Turbo refinement pass. The source video says the secret is using the Krea 2 Turbo LoRA/rank LoRA with the Turbo model instead of the raw model, producing a light detail pass in only a few additional steps.
Conceptual graph
Prompt + Krea 2 Turbo settings
│
▼
First Krea 2 generation ──► save/preview baseline image
│
▼
Second light Krea 2 Turbo pass with Turbo LoRA/refiner settings
│
▼
Refined output with more texture, sharpness, and detail
Practical setup steps
- Keep the first pass normal. Use your known-working Krea 2 model, encoder, VAE, sampler, and prompt settings.
- Route the first image into the second pass. The second pass should use the first output as its starting image/reference, not create an unrelated composition.
- Use a light second pass. The video describes the combo as preserving the original composition while adding detail in roughly four extra steps. Start with low steps/low denoise rather than a heavy regeneration.
- Use the Turbo LoRA/refinement setup exactly as the workflow expects. If the workflow includes a Turbo LoRA loader, select the correct Krea 2 Turbo LoRA and model pairing.
- Save both outputs. Keep the first-pass image and the refined image. Sometimes the unrefined version has better composition or softer aesthetics.
- Compare side by side. Look for sharper textures, better edges, and detail gains, but also check for over-sharpening, busy backgrounds, warped text, or face/hand degradation.
| Second-pass setting | Starting recommendation | Adjustment logic |
|---|
| Steps | Low, around the workflow’s intended light-refine value; the video mentions four extra steps. | Increase only if the second pass is too weak; lower if it overcooks textures. |
| Denoise / strength | Low enough to preserve composition. | If the image changes identity or layout, reduce denoise/strength. |
| Prompt | Reuse the first-pass prompt, or keep the second-pass prompt minimal. | The video recommends not loading extra LoRAs into the second pass; let it refine, not re-style. |
| LoRAs | Use style LoRAs in the first pass, not the second pass. | Putting style adapters in both passes can double-apply effects and drift the image. |
| Seed | Fixed for testing. | Use random only after you know the combo settings work. |
6. Use the IdeoKrea LoRA
Aitrepreneur’s IdeoKrea LoRA is an experimental Krea 2 LoRA intended to make Krea 2 outputs resemble Ideogram 4’s sharper, more contrasty, poster-like style. The video shows it improving color pop, contrast, and style in several examples.
Installation
- Download the IdeoKrea LoRA from Hugging Face.
- Place the LoRA file in
ComfyUI/models/loras/.
- Refresh model lists with R or restart ComfyUI.
- Select it in your LoRA loader.
- Start with moderate strength and a fixed seed.
How to use it in the combo workflow
- Use IdeoKrea in the first pass. The presenter recommends applying LoRAs to the first generation so the second pass refines the resulting image consistently.
- Leave the second pass clean. Avoid adding the IdeoKrea LoRA again in the second pass unless you specifically want an exaggerated style test.
- Try both prompt styles. The video demonstrates using an Ideogram-style prompt builder, but also says normal natural-language prompting can work because the underlying model is still Krea 2.
- Expect it to be experimental. The presenter describes it as a test version, so keep it as an option, not a default for every image.
7. Recommended comparison settings
Use this test plan to decide whether each new feature helps your actual images.
| Run | Enhancer | IdeoKrea | 2× combo | What it tells you |
|---|
| Baseline | Off | Off | Off | What plain Krea 2 does with your prompt. |
| Enhancer only | 0.1 | Off | Off | Whether prompt following improves without style damage. |
| Combo only | Off | Off | On | Whether the second pass adds useful detail. |
| IdeoKrea only | Off | On | Off | Whether you like the contrast/style shift. |
| Full stack | Low | First pass only | On | Whether all enhancements cooperate without over-styling. |
Keep a visual contact sheet: save all five outputs with clear names such as baseline_seed123.png, enhancer01_seed123.png, and combo_seed123.png. This prevents “newer looks better” bias.
8. Troubleshooting
“The enhancer node does not appear.”
- Confirm the custom node folder is inside
ComfyUI/custom_nodes/, not nested one level too deep.
- Restart ComfyUI after installing; browser refresh alone is often not enough.
- Read the ComfyUI terminal for Python import errors or missing package messages.
- If the node/package name changed, search the node palette for
Krea, Krea2T, Enhancer, and Conditioning.
“The enhancer changes the style too much.”
- Reduce strength toward
0.05–0.1.
- Verify the seed, sampler, and model are identical to the baseline.
- Use the enhancer only on prompts where a specific detail is failing.
“The second pass makes the image too sharp or weird.”
- Lower second-pass steps or denoise/strength.
- Remove extra LoRAs from the second pass.
- Use the first-pass image as the final if the refined version adds artifacts.
- Check whether your workflow accidentally reruns text-to-image instead of image refinement.
“LoRA has no visible effect.”
- Confirm the LoRA file is in
models/loras/ and appears in the dropdown.
- Check whether the LoRA requires a trigger word.
- Increase strength gradually and compare with a fixed seed.
- Confirm the LoRA was trained for Krea 2 or is compatible with the model architecture you are using.
“I get model/text-encoder shape errors.”
- Make sure the Krea model, Qwen text encoder, and VAE are the exact files expected by the workflow.
- Do not mix Qwen Image Edit encoders, unrelated CLIP files, and Krea 2 loaders at random.
- Return to a known-working baseline workflow, then add the combo nodes again one at a time.
9. Verification checklist
- ComfyUI launches without custom-node import errors.
- Baseline Krea 2 image generation works with all enhancements disabled.
- Enhancer node appears and produces a completed image at low strength.
- First-pass and second-pass images are both saved so you can compare them.
- The second pass improves detail without changing composition more than intended.
- IdeoKrea appears in the LoRA dropdown and changes style when enabled.
- Style LoRAs are applied in the first pass only unless you intentionally test otherwise.
- Outputs are reviewed for legal/safety boundaries before sharing or publishing.
10. Sources and useful links
Prepared from the public YouTube transcript and metadata on 2026-06-27. Links were checked during drafting, but model files, node names, licenses, and ComfyUI workflows can change quickly; verify official pages before production use.