h4 - Gridinator
- optional_vae
- image_input
- Grid_Image
◄image_upload▾►
◄base_model▾►
◄base_model_fuzzy►
◄width1024►
◄height1024►
◄batch_size1►
◄positive_promptAn epic photo of...►
◄negative_promptblurry, low quality►
◄seed0►
◄steps20►
◄cfg7.0►
◄sampler_name▾►
◄scheduler▾►
◄denoise1.00►
◄lora_strength1.00►
◄grid_x_modeNone►
◄grid_x_val►
◄grid_x_override►
◄grid_y_modeNone►
◄grid_y_val►
◄grid_y_override►
◄grid_z_modeNone►
◄grid_z_val►
◄grid_z_override►
◄stutter_modeOff►
◄sliding_scale_enablefalse►
◄denoise_min0.20►
◄denoise_max0.80►
◄steps_min10►
◄steps_max30►
◄range_count4►
◄font_size40►
◄font_colorwhite►
◄bg_colorblack►
◄margin50►
◄padding20►
Categoryh4_Live/Grid
Inputs (38)
| Name | Type | Default | Description |
|---|---|---|---|
| image_upload | COMBO | CRITICAL FOR IMG2IMG: Upload the base image you want me to modify here. This image acts as the 'source' for all my grid variations. If you leave this empty and haven't connected an external image, I will default to Txt2Img mode automatically. Supported formats: .jpg, .png, .webp. Tip: If your grid looks like colorful noise, check if your Denoise is too high (1.0) or your image didn't load! | |
| base_model | COMBO | THE BRAIN: Choose your primary Checkpoint/Model here. This is the foundation from which I will draw all images. For grids comparing OTHER models (Architecture mode), this acts as my 'Default' model if an axis value is missing or for any axis not designated as 'Model'. Note: SDXL models require you to use 1024x1024, SD1.5 models prefer 512x512. | |
| base_model_fuzzy | STRING | FUZZY HUNTER: Can't find your model in the dropdown? Type a fragment of the name here (e.g., 'juggernaut' or 'pony'). I will scan your entire models folder and find the best match for you. This is great for those with 1000+ models where dropdowns become a nightmare. If filled, this OVERRIDES my dropdown selection above. | |
| width | INT | 102464–8192 | HORIZONTAL RESOLUTION: The width of your output images in pixels. Rule of Thumb: 512 for SD1.5, 768 for SD2.1, 1024 for SDXL. Going too high here without a powerful GPU will cause 'Out of Memory' (OOM) errors. Always ask me for multiples of 8! |
| height | INT | 102464–8192 | VERTICAL RESOLUTION: The height of your output images in pixels. Tall for portraits (1216), Square for general (1024), Wide for landscapes (768/832). Same memory rules as Width apply. High resolutions exponentially increase my generation time and your VRAM usage. |
| batch_size | INT | 11–64 | INTERNAL BATCH: How many images you want me to generate SIMULTANEOUSLY for a single cell. Usually set to 1 for grids to keep my layout clean. If you set this higher, the grid will only show the FIRST image of the batch, which is usually a waste of VRAM unless you are doing complex batch-processing. |
| positive_prompt | STRING | An epic photo of... | DREAM DICTATION: Tell me what you want to see. Be descriptive! Use commas to separate your concepts. Use [weight*1.2] to emphasize words to me. If you use the 'Prompt' Axis Mode, THIS BOX WILL AUTOMATICALLY HIDE to prevent confusion, as your axis values will take over the driver's seat. Tip: Start with subject, then lighting, then style, then camera details. |
| negative_prompt | STRING | blurry, low quality | THE FORBIDDEN ZONE: Tell me what you DON'T want in your art. List things like 'deformed iris', 'bad hands', 'watermark', 'text'. A strong negative prompt is the secret sauce to me giving you professional-looking art. Most models have specific recommended negatives (e.g., 'lowres, bad anatomy'). |
| seed | INT | 00–18446744073709550000 | THE COSMIC RNG: The starting point of my noise generation. 0 means every run is unique (Random). Using a fixed number allows me to reproduce the EXACT same image for you if nothing else changes. Think of it as the DNA of the generation. In a grid, if Seed isn't an axis, every cell uses this same DNA to ensure differences you see are from the parameters, not luck. |
| steps | INT | 201–100 | THOUGHT CYCLES: How many times you want me to refine the image. 20 is the sweet spot for speed/quality. 1-10 results in a blurry mess. 30-50 adds fine detail for you but takes me longer. Above 50 often hits diminishing returns or introduces artifacts. High step counts = High time cost. |
| cfg | FLOAT | 7.00–100 | CREATIVITY VS OBEDIENCE: Classifier Free Guidance. 1.0 = I completely ignore you. 7.0 = Standard balance. 10.0-15.0 = I follow your words extremely strictly but I might 'fry' the colors and edges. Think of it as my 'Volume' knob. |
| sampler_name | COMBO | THE MATHEMATICIAN: The algorithm I use to denoise. 'euler' is the classic. 'dpmpp_2m' or 'dpmpp_sde' with 'karras' are the modern champions for realism. Each sampler has a different 'personality' - experiment to find your favorite look! | |
| scheduler | COMBO | THE PACEMAKER: Controls how I remove the noise over the steps. 'normal' is linear. 'karras' focuses on early/late steps for better convergence. 'simple' is fast. 'exponential' is aggressive. Usually paired with specific samplers recommended by model creators. | |
| denoise | FLOAT | 1.000–1 | RECONSTRUCTION STRENGTH: 1.0 = Total replacement (Standard Txt2Img). 0.5 = I keep 50% of your source image and add 50% generation. 0.0 = Pixel-perfect match (I make no change). ESSENTIAL for Img2Img - usually start around 0.6 to keep your base structure while I add detail. |
| lora_strength | FLOAT | 1.00-10–10 | STYLE AMPLIFIER: How much of the chosen LoRA's personality you want me to inject. 1.0 is full effect. -1.0 reverses the effect (weird!). Be careful: high strengths (above 1.5) can destroy the image coherence. Ask me for 0.7-0.9 for subtle blending. |
| grid_x_mode | COMBO | None | X-AXIS (COLUMNS): Choose which parameter varies from Left to Right. If you set this to 'Prompt', I will split your input by NEWLINES. If 'Multi-Prompt', I will split by SEMICOLONS (;). These modes preserve your commas, allowing individual complex prompts! |
| grid_x_val | STRING | AXIS VALUES: Enter your variants here. For 'Prompt' mode, separate with NEWLINES. For 'Multi-Prompt', separate with SEMICOLONS (;). For standard modes (Steps, CFG), use COMMAS. If you pick 'Model' or 'LoRA', use the dropdown or right-click to pick files! | |
| grid_x_override | STRING | MANUAL OVERRIDE: If you want to use files/values NOT in the dropdowns (like a subfolder model), type them here. This takes absolute priority over the Mode/Value boxes above. Format: 'filenameA, filenameB'. Commas are the separator here. | |
| grid_y_mode | COMBO | None | Y-AXIS (ROWS): Choose which parameter varies from Top to Bottom. Tip: Compare Samplers on X and Schedulers on Y for a masterclass in algorithm differences. If using a prompt mode, it will override the base prompt below! |
| grid_y_val | STRING | AXIS VALUES: Same as X-Axis. Remember: Prompt = Newlines, Multi-Prompt = Semicolons (;), Others = Commas. | |
| grid_y_override | STRING | MANUAL OVERRIDE: Same as X-Axis override. Takes priority over Y-Axis boxes. Commas are ALWAYS the separator for manual overrides. | |
| grid_z_mode | COMBO | None | Z-AXIS (STACKS): The 3rd Dimension. This creates multiple distinct grids or a 'deep stack'. Use this for comparing across Models while X/Y compare Prompts/Settings. If set to a prompt mode, each prompt creates a new stack! |
| grid_z_val | STRING | AXIS VALUES: Same as X/Y. Remember: Prompt = Newlines, Multi-Prompt = Semicolons (;), Others = Commas. | |
| grid_z_override | STRING | MANUAL OVERRIDE: Same as X/Y override. Takes priority over Z-Axis boxes. Commas are ALWAYS the separator for manual overrides. | |
| stutter_mode | COMBO | Off | PROMPT AUTOMATION: 'Permutations' splits {red|blue} into multiple cells (if your axis mode matches). 'Emphasis' expands [dog*3] into 'dog dog dog'. A massive time-saver for repetitive prompt testing. |
| sliding_scale_enable | BOOLEAN | false | AUTO-GENERATOR: Tired of typing '10, 20, 30, 40'? Turning this ON allows me to automatically generate a linear range between the Min and Max values below. This only applies if you set an Axis Mode to 'Steps' or 'Denoise'. |
| denoise_min | FLOAT | 0.200–1 | RANGE START (Denoise): The lowest value in your generated scale. |
| denoise_max | FLOAT | 0.800–1 | RANGE END (Denoise): The highest value in your generated scale. |
| steps_min | INT | 101–100 | RANGE START (Steps): The lowest step count in your generated scale. |
| steps_max | INT | 301–100 | RANGE END (Steps): The highest step count in your generated scale. |
| range_count | INT | 42–100 | DENSITY: How many images you want me to spread across the range. e.g. Start 10, End 40, Count 4 means I will give you [10, 20, 30, 40]. |
| font_size | INT | 4010–200 | LEGIBILITY: The point size of the grid labels I draw for you. If your grid is huge, make this larger so you can read it in the preview! |
| font_color | STRING | white | AESTHETICS: The color of the label text I render. I accept standard names like 'red', 'gold', 'lime' or Hex codes like '#FF00FF'. |
| bg_color | STRING | black | CANVAS COLOR: The color of the space between images and the outer margins. Dark colors usually make images 'pop' better. |
| margin | INT | 500–500 | OUTER SPACE: The thickness of the border around the entire final grid sheet in pixels. |
| padding | INT | 200–200 | INNER SPACE: The gap between individual image cells and between images and their labels. |
| optional_vaeopt | VAE | VAE OVERRIDE: Optional. If you want me to use a specific VAE instead of the one baked into the model, connect it here. This is critical for SDXL models if your base checkpoint has a broken VAE or for specialized artistic looks. | |
| image_inputopt | IMAGE | EXTERNAL IMAGE: If you have an image coming from another node (like a Load Image or a Masked image), connect it here. I will prioritize this over your 'image_upload' dropdown above. Essential for complex Img2Img pipelines! |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| Grid_Image | IMAGE | — |