Free and Full
The Free edition has no limit that stops you from making an excellent image: same render, same Detail boost, same face pass and upscale. The Full edition makes the workflow faster and nicer to use every day, and adds the photographic finishing tools.
| Feature | Free | Full |
|---|---|---|
| ⓪ Start: one-click downloads, help | ✓ | ✓ |
| Prompt, 10 styles, formats, exact custom size | ✓ | ✓ |
| Render with Detail boost and live preview | ✓ | ✓ |
| Face pass, tiled upscale, step previews | ✓ | ✓ |
| Save as PNG, WebP or JPEG, Civitai metadata, before and after, lightbox | ✓ | ✓ |
| LoRAs | 3, from a list | Unlimited |
| Visual LoRA library: previews, base-model filter, ★ Characters, custom covers | — | ✓ |
| Natural look, Camera finish, Film look, Adjustments | — | ✓ |
| Shot presets: lens, light, framing | — | ✓ |
| Signature with text or your logo | — | ✓ |
| Social formats framed on the faces | — | ✓ |
| History with restore, presets, Draft and Final | — | ✓ |
Both editions use the same folder and the same nodes. Upgrading is replacing the folder, and a workflow saved in one edition opens in the other. The Full edition is on Ko-Fi for $9.99.
Install
- Unzip into custom_nodesPut the
ComfyUI-Enzino-Krea2-WFfolder inComfyUI/custom_nodes/. - Keep one copy onlyFree and Full use the same folder name. To keep the other edition around, rename its folder to
ComfyUI-Enzino-Krea2-WF.disabled: ComfyUI skips folders ending in.disabled. - Restart ComfyUI fullyClose it and start it again, then reload the page with Ctrl + F5. Reloading the page alone doesn't load new node code.
- Open the workflowDrag
workflows/enzino_krea2_wf.jsononto ComfyUI, then follow the ⓪ Start block.
The console confirms what is loaded:
If the console also shows ANOTHER COPY IS ALREADY INSTALLED, two copies are loaded and ComfyUI may run the old one. Remove or disable one of them and restart.
Needs a recent ComfyUI with native Krea 2 support. The face pass uses ultralytics, which Impact Pack already installs; otherwise run pip install ultralytics in ComfyUI's Python.
⓪ Start: downloads and help
The Start block checks what the workflow needs and downloads the missing files in one click, straight into the right ComfyUI folder, with progress, pause and resume. Downloads only come from a fixed list of sources. Once everything is installed, the list folds into Ready to create.
| Need | Choices |
|---|---|
| Krea 2 model | Any Krea 2 model works. IntoRealism Krea 2 · Flash (4 steps) is tuned for this workflow. Krea 2 Turbo fp8 (13.1 GB), nvfp4 for RTX 50 cards (7.7 GB) or bf16 (26.3 GB). |
| Text encoder | Qwen3-VL 4B, fp8 (5.2 GB) or bf16 (8.9 GB) |
| VAE | Krea 2 VAE (254 MB). The Wan 2.1 VAE works too. |
| Upscaler optional | 1x-ITF-SkinDiffDetail-Lite (20 MB, natural skin detail) or 4x-UltraSharp (67 MB, sharper) |
| Face detector optional | face_yolov8m (52 MB). It also downloads itself the first time the face pass runs. |
IntoRealism in one click
Civitai only serves downloads to signed-in users. Click 🔑 next to IntoRealism and paste your free Civitai API key once. You create it on civitai.com/user/account, under API Keys. The key stays on your machine, in ComfyUI/user/default/enzino_krea2_pipeline/, and is never shown again. Without a key, the Civitai button opens the model page.
The six nodes
Every node passes one bundle of data to the next on a single wire. Each card shows the main settings; the rest sits under Advanced ▸. Each node also exposes the usual ComfyUI outputs (model, clip, image…) if you want to branch off.
1Models
The Krea 2 model, the Qwen3-VL text encoder and the VAE in one node. It loads exactly the files you pick. A file saved in a workflow but missing on your machine is flagged in red, with a pointer to ⓪ Start.
Advanced holds the weight type: fp8 saves VRAM on 12–16 GB cards.
2LoRAs
Each LoRA gets a strength slider, an on and off switch and its trigger words as chips. Trigger words come from the LoRA's .civitai.info file or its metadata, and the active ones are added to the prompt for you. A file named my_lora.preview.png next to the LoRA becomes its thumbnail.
Free: up to 3 LoRAs, picked from a list sorted by folder.
Full FULL: unlimited LoRAs and the visual library below.
The visual library FULL
- Search, folders and base-model filters: the base model (Krea 2, Z-Image, Qwen Image, Flux, SDXL, SD 1.5, Wan) is read from each file, so mixed folders stay usable.
- ★ Characters: click ＋ Add a character and pick a LoRA, or click ☆ on any card. With Enzino Krea2 Studio Pro installed, the characters of your Cast gallery show up here too.
- Custom covers: hover a card and click 🖼 to choose an image, or to go back to the original. Covers are stored in your user folder, never next to the LoRAs, and show everywhere: library, LoRA node, Characters.
- Click a card to add it, click again to remove it, Esc to close.
3Prompt
Describe the photo; the style and the LoRA trigger words are added to it. Advanced holds the negative prompt and where trigger words go (start, end or off).
| Styles | Analog film, Editorial, Cinematic, Studio portrait, Smartphone candid, Golden hour, Night street, Black & white, Vintage 90s, or none |
| Format | 1:1, 4:5, 3:4, 2:3, 9:16, 16:9, 3:2, 21:9 with a size in megapixels, plus a batch size |
| Custom size | Any width and height, exact to the pixel (1500 × 2400 gives 1500 × 2400). The image renders on the next multiple of 16, then is cropped back. Quick sizes: 1080 × 1350, 1080 × 1920, 1500 × 2400… |
| Shot FULL | Lens (24 mm to 135 mm, macro, smartphone), light (window, golden hour, overcast, noon sun, flash, neon, softbox, candlelight) and framing (close-up to wide scene) |
Krea 2 Turbo is at its best around 1.5–2 MP. Above about 2.5 MP it can repeat patterns, and the node shows a warning: render smaller and let ⑤ Upscale do the rest.
4Render
The preview shows the image forming while it samples, then the raw render. or a double-click opens it in the lightbox.
| Seed | 🎲 for a new random seed, 🔒 to keep the same seed between runs |
| Steps | IntoRealism Flash: 4. Krea 2 Turbo: 8. |
| Detail boost | Keeps skin, hair and fabric texture instead of smoothing it. 0.3 by default; 0.2–0.5 on Turbo and Flash, 0 to turn it off. It only nudges the middle steps, never the first one (composition) or the last one (clean-up). |
| Advanced | CFG (1), sampler (euler), scheduler (simple) |
5Enhance
A grid of steps: click a tile to see its settings below, click the dot to turn it on or off. The preview strip shows the image after each step; click a thumbnail to view it, to open the lightbox and compare with ← →. Changing a later step only reruns that step: tweaking the film look doesn't redo the faces or the upscale.
| Faces | Every face is found, refined at 1024 px and blended back with its original tones. Strength 0.10 with 1 step by default; above 0.2 Turbo starts adding moles and new details. Prompt: natural skin texture, subtle pores. |
| Upscale | Upscale model, then a tiled re-detail pass with soft seams and colour matching. ×1.5, added detail 0.15, 2 steps, sharpen 0.2 by default. Tiles: 1152 px, overlap 192. |
| Natural look FULL | Calms loud colours, natural skin tones, soft highlights, fine texture: removes the glossy generated look while keeping contrast and deep blacks |
| Camera FULL | The sensor and lens signature of an iPhone, Fuji X100, 50 mm DSLR, 2000s compact, disposable flash or Android night shot |
| Film FULL | Clean, Kodak Portra, Cinestill 800T, Fuji 400H, Tri-X B&W, Moody, Teal & Orange, with grain, vignette and halation |
| Adjustments FULL | Brightness, contrast, saturation and grain, plus highlights, shadows, vibrance, clarity, temperature, tint, sharpness and vignette under Advanced. Applied last, 0 = unchanged, one click to reset. |
| Signature FULL | Text, your logo or both, in six positions. Drop a PNG with transparency on the card to use it as a logo. |
6Output
- Before and after: move the mouse over the large image to compare the raw render with the final one.
- Final image card: View opens the lightbox (click to zoom to 100 %), New tab opens the file, ⬇ downloads it.
- Save: folder inside
outputwith a folder browser, file name, name parts (counter, date, time, size, batch number), PNG, WebP or JPEG. The ComfyUI workflow and Civitai metadata (prompt, LoRAs, seed, steps) can be written into the file, so Civitai fills in the resources when you post. - Social formats FULL: 4:5 (1080 × 1350), 9:16 (1080 × 1920), 1:1 (1080 × 1080) and 16:9 (1920 × 1080), framed on the faces, signed if the signature is on, saved in
output/social. - History FULL: your recent renders. Click one to restore all its settings or just its seed.
- Optional image input: plug any image node between ⑤ Enhance and ⑥ Output, such as a filter or a neural renderer. Connect ⑤'s
imageoutput to it and its result to ⑥'simageinput; that image is the one saved and shown.
Header, Draft and presets FULL
- ⚡ Draft / ✦ Final: Draft turns off Faces, Upscale, Film and Social formats to explore prompts and seeds fast. Final turns your settings back on.
- Presets: Portrait · IntoRealism, Editorial, Cinematic night, Black & white, Smartphone candid, Fuji street and Fast draft, plus your own. ＋ saves the current settings of every node under a name.
Recommended settings
These are the defaults. Turbo and Flash models are sensitive: small values go a long way.
| Setting | IntoRealism Flash | Krea 2 Turbo |
|---|---|---|
| Steps · CFG · sampler | 4 · 1 · euler / simple | 8 · 1 · euler / simple |
| Size | 1.5–2 MP, then Upscale ×1.5 | |
| Detail boost | 0.3 (0.2–0.5) | |
| Faces | 0.10 with 1 step (0.08–0.2) | |
| Upscale added detail | 0.15 with 2 steps (0.1–0.2), SkinDiffDetail upscaler | |
Troubleshooting
Generation is slow
The console prints the time of each step:
The first run loads the models from disk. Later runs should be much faster, and settings that only change a later step reuse the earlier ones. If ComfyUI writes loaded partially, your graphics card is short on memory: use fp8 in ① Advanced or a lighter model. Also check that only one copy of the pack is installed (see Install).
① Models is outlined in red
The workflow refers to a file name you don't have. Pick your own file in the menu; any Krea 2 model works.
Faces get moles or new details
Lower Faces to 0.08–0.15 and keep 1 step. Avoid words like detailed face or sharp eyes in the face prompt: Turbo takes them literally.
Odd texture, droplets or repeated patterns
Lower the Detail boost, or render smaller (1.5–2 MP) and upscale.
No face is refined
The face pass needs ultralytics (installed by Impact Pack, or pip install ultralytics) and a face model, which ⓪ Start downloads. Lower Detection in Faces ▸ Advanced for small or turned faces.
Previews are gone after a restart
Step previews are temporary files that ComfyUI clears when it restarts. Saved images stay in output. Switching workflow tabs keeps the previews.
Works with Enzino Krea2 Studio
Both packs install side by side. Krea2 Studio puts several characters, each with their own LoRA, in one Krea 2 image. With Krea2 Studio Pro installed, the characters of its Cast gallery appear in the ★ Characters tab of the LoRA library, with their cover photo.
The nodes are Enzino's code. Krea 2 and its derivatives, including IntoRealism Krea 2, are under the Krea 2 Community License; outputs and redistributed models must follow it. The bundled Inter font is under the SIL Open Font License.