Merge pull request #1 from Visionatrix/sd3-flow

added SD3-Medium model/flow

README.md

@@ -5,3 +5,11 @@ This repository is intended only for Workflows - if you want to add a flow, make
 The Visionatrix documentation is also stored here to not overload the main repository.
 
 For any problems with Visionatrix or suggestions for improvement, go to the [main repository](https://github.com/Visionatrix/Visionatrix).
+
+- [Documentation](https://visionatrix.github.io/VixFlowsDocs/)
+  - [Available Flows](https://visionatrix.github.io/VixFlowsDocs/Flows/index.html)
+  - [Manual Installation](https://visionatrix.github.io/VixFlowsDocs/Installation.html)
+  - [Command Line Options](https://visionatrix.github.io/VixFlowsDocs/CommandLineOptions.html)
+  - [Working modes](https://visionatrix.github.io/VixFlowsDocs/WorkingModes.html)
+  - [Vix Workflows](https://visionatrix.github.io/VixFlowsDocs/VixWorkflows.html)
+  - [Technical information](https://visionatrix.github.io/VixFlowsDocs/TechnicalInformation.html)

docs/Flows/SD3_Medium.rst

@@ -0,0 +1,39 @@
+.. _SD3_Medium:
+
+StableDiffusion3-Medium
+=======================
+
+Flow using the gated model, requires a HuggingFace token to setup.
+
+*This flow is made more for development and verification that we have successfully added the ability to use gated/closed models.*
+
+.. note::
+    In the future this Flow will either be modernized or removed when something better appears to replace it based on the feature-tuned SD3, if there are any.
+
+**Supports various aspect ratios.**
+
+Hardware
+""""""""
+
+- **Required memory: 12-16 GB**
+
+Time to generate 1 image:
+
+- AMD 7900 XTX: **22.1 sec**
+- NVIDIA RTX 3060 (12 GB): **25-32 sec**
+- Apple M2 Max: **97 sec**
+
+Examples
+""""""""
+
+.. image:: /FlowsResults/SD3_Medium_1.png
+
+Prompt: "*Black kitten with white wings sitting on a blue cloud, cinematic*" | prompt_strength: 5.1 (seed: 6814591)
+
+.. image:: /FlowsResults/SD3_Medium_2.png
+
+Prompt: "*poster, cyborg girl against an alien, black baground, high contrast, high details, cinematic*" | prompt_strength: 5.5 (seed: 2131028)
+
+.. image:: /FlowsResults/SD3_Medium_3.png
+
+Prompt: "*an oil line art painting of the angel impressive neon shadows, warm colors*" | prompt_strength: 4.1 (seed: 1167357)

docs/Flows/index.rst

@@ -18,3 +18,4 @@ Available Flows
    ComicuPortrait
    MemojiPortrait
    SupirUpscaler
+   SD3_Medium

docs/GatedModels.rst

@@ -0,0 +1,29 @@
+Gated Models
+============
+
+It often happens that the model you are using is not available for download without authentication. These are referred to as `Gated Models <https://huggingface.co/docs/hub/models-gated>`_.
+
+Flows with such models have a separate mark in the Visionatrix UI.
+
+To be able to install such a flow, you need to specify an ``Access Token``
+
+.. note::
+    Currently, only HuggingFace Access Tokens are supported.
+
+Steps to Access Gated Models:
+
+1. Register on `HuggingFace <https://huggingface.co>`_ if you are not already registered
+2. Gain access to the model on your account by going to its page (you can click on the model from Visionatrix UI) and filling out the form
+3. Generate an access token in the settings of HuggingFace (click on your icon -> settings -> access tokens)
+4. Click on ``Set Permissions`` of the token after generation and select ``Read access to contents of all public gated repos you can access``
+5. Go to the Visionatrix settings and enter this access token
+
+Alternatively, you can set an environment variable named ``HF_AUTH_TOKEN`` with the token value, but this requires setting up the environment variable for each worker if you have many of them.
+
+I'm a user and want to connect my own worker to process flows with closed models.
+---------------------------------------------------------------------------------
+
+As user's workers cannot receive global access tokens from the server to avoid leaks, you have two options:
+
+1. Download the model yourself and place it in the folder specified in ``models_catalog.json`` under the ``save_path`` key.
+2. Set the ``HF_AUTH_TOKEN`` environment variable with your own public access token, and the worker will be able to install flows with gated models.

docs/index.rst

@@ -10,6 +10,7 @@ Here will leave all docs that is not suitable for Readme file.
     Installation.rst
     CommandLineOptions.rst
     WorkingModes.rst
+    GatedModels.rst
     VixWorkflows.rst
     TechnicalInformation.rst
 

flows/sd3_medium/flow.json

@@ -0,0 +1,135 @@
+{
+  "name": "sd3_medium",
+  "display_name": "StableDiffusion3 Medium",
+  "description": "Simple one model flow for testing purposes",
+  "author": "Datou",
+  "homepage": "https://openart.ai/workflows/datou/sd3-basic-workflow/7Udt2JZ3Mwqf1if9EqWt",
+  "license": "",
+  "documentation": "https://visionatrix.github.io/VixFlowsDocs/Flows/SD3_Medium.html",
+  "tags": ["general"],
+  "input_params": [
+    {
+      "name": "prompt",
+      "display_name": "Prompt",
+      "type": "text",
+      "optional": false,
+      "comfy_node_id": {
+        "6": {
+          "dest_field_name": [
+            "inputs",
+            "text"
+          ]
+        }
+      }
+    },
+    {
+      "name": "negative_prompt",
+      "display_name": "Negative Prompt",
+      "type": "text",
+      "default": "bad quality, poor quality, doll, disfigured, jpg, toy, bad anatomy, missing limbs, missing fingers, 3d, cgi",
+      "optional": false,
+      "advanced": true,
+      "comfy_node_id": {
+        "71": {
+          "dest_field_name": [
+            "inputs",
+            "text"
+          ]
+        }
+      }
+    },
+    {
+      "name": "aspect_ratio",
+      "display_name": "Aspect Ratio",
+      "type": "list",
+      "optional": true,
+      "advanced": true,
+      "default": "1x1 (1024x1024)",
+      "options": {
+        "1x1 (1024x1024)": [
+          1024,
+          1024
+        ],
+        "2x3 (832x1216)": [
+          832,
+          1216
+        ],
+        "3x4 (896x1152)": [
+          896,
+          1152
+        ],
+        "5:8 (768x1216)": [
+          768,
+          1216
+        ],
+        "9:16 (768x1344)": [
+          768,
+          1344
+        ],
+        "9:19 (704x1472)": [
+          704,
+          1472
+        ],
+        "9:21 (640x1536)": [
+          640,
+          1536
+        ],
+        "3:2 (1216x832)": [
+          1216,
+          832
+        ],
+        "4:3 (1152x896)": [
+          1152,
+          896
+        ],
+        "8:5 (1216x768)": [
+          1216,
+          768
+        ],
+        "16:9 (1344x768)": [
+          1344,
+          768
+        ],
+        "19:9 (1472x704)": [
+          1472,
+          704
+        ],
+        "21:9 (1536x640)": [
+          1536,
+          640
+        ]
+      },
+      "comfy_node_id": {
+        "135": {
+          "dest_field_name": [
+            "inputs",
+            [
+              "width",
+              "height"
+            ]
+          ]
+        }
+      }
+    },
+    {
+      "name": "prompt_strength",
+      "display_name": "Prompt strength",
+      "type": "range",
+      "optional": true,
+      "advanced": true,
+      "default": 4.1,
+      "min": 4.0,
+      "max": 7.0,
+      "step": 0.1,
+      "comfy_node_id": {
+        "271": {
+          "dest_field_name": [
+            "inputs",
+            "cfg"
+          ],
+          "internal_type": "float"
+        }
+      }
+    }
+  ]
+}

flows/sd3_medium/flow_comfy.json

@@ -0,0 +1,107 @@
+{
+  "6": {
+    "inputs": {
+      "text": "",
+      "clip": [
+        "252",
+        1
+      ]
+    },
+    "class_type": "CLIPTextEncode",
+    "_meta": {
+      "title": "CLIP Text Encode (Prompt)"
+    }
+  },
+  "71": {
+    "inputs": {
+      "text": "bad quality, poor quality, doll, disfigured, jpg, toy, bad anatomy, missing limbs, missing fingers, 3d, cgi",
+      "clip": [
+        "252",
+        1
+      ]
+    },
+    "class_type": "CLIPTextEncode",
+    "_meta": {
+      "title": "CLIP Text Encode (Negative Prompt)"
+    }
+  },
+  "135": {
+    "inputs": {
+      "width": 1024,
+      "height": 1024,
+      "batch_size": 1
+    },
+    "class_type": "EmptySD3LatentImage",
+    "_meta": {
+      "title": "EmptySD3LatentImage"
+    }
+  },
+  "231": {
+    "inputs": {
+      "samples": [
+        "271",
+        0
+      ],
+      "vae": [
+        "252",
+        2
+      ]
+    },
+    "class_type": "VAEDecode",
+    "_meta": {
+      "title": "VAE Decode"
+    }
+  },
+  "252": {
+    "inputs": {
+      "ckpt_name": "sd3_medium_incl_clips_t5xxlfp16.safetensors"
+    },
+    "class_type": "CheckpointLoaderSimple",
+    "_meta": {
+      "title": "Load Checkpoint"
+    }
+  },
+  "271": {
+    "inputs": {
+      "seed": 71844603528225,
+      "steps": 28,
+      "cfg": 4.1,
+      "sampler_name": "dpmpp_2m",
+      "scheduler": "sgm_uniform",
+      "denoise": 1,
+      "model": [
+        "252",
+        0
+      ],
+      "positive": [
+        "6",
+        0
+      ],
+      "negative": [
+        "71",
+        0
+      ],
+      "latent_image": [
+        "135",
+        0
+      ]
+    },
+    "class_type": "KSampler",
+    "_meta": {
+      "title": "KSampler"
+    }
+  },
+  "273": {
+    "inputs": {
+      "filename_prefix": "comfy_",
+      "images": [
+        "231",
+        0
+      ]
+    },
+    "class_type": "SaveImage",
+    "_meta": {
+      "title": "@Datou"
+    }
+  }
+}

models_catalog.json

@@ -417,5 +417,18 @@
         "input_name": "^(?!.*(?i:lora)).*"
       }
     ]
+  },
+  "SD3-Medium" : {
+    "save_path": "checkpoints/sd3_medium_incl_clips_t5xxlfp16.safetensors",
+    "url": "https://huggingface.co/stabilityai/stable-diffusion-3-medium/resolve/main/sd3_medium_incl_clips_t5xxlfp16.safetensors",
+    "homepage": "https://huggingface.co/stabilityai/stable-diffusion-3-medium",
+    "hash": "69a950c5d143ce782a7423c532c8a12b75da6a37b0e6f26a322acf4e76208912",
+    "regexes": [
+      {
+        "input_value": "^(?=.*(?i:sd3))(?=.*(?i:medium))(?=.*(?i:fp16)).*",
+        "input_name": "^(?!.*(?i:lora)).*"
+      }
+    ],
+    "gated": true
   }
 }