# ISIS v2 Character Concept Art Control-Conditioned Workflow Graph

Generated: 2026-02-24T21:13:07.031Z Checklist: `MXA-0004` (ISIS2-00202) Asset
Family: `character-concept-art` Phase: Phase 4 | Milestone: M3 Launch
(2026-07-27 to 2026-10-30)

## Objective

Define a control-conditioned workflow graph for character concept art that
supports pose/depth/segmentation constraints while preserving deterministic
defaults.

## Deterministic Defaults

| Setting                    | Value                                    |
| -------------------------- | ---------------------------------------- |
| `checkpointModel`          | `sd_xl_base_1.0.safetensors`             |
| `seed`                     | `2026072704`                             |
| `width`                    | `1024`                                   |
| `height`                   | `1536`                                   |
| `steps`                    | `35`                                     |
| `cfgScale`                 | `6.4`                                    |
| `samplerName`              | `dpmpp_2m_sde`                           |
| `scheduler`                | `karras`                                 |
| `denoise`                  | `1`                                      |
| `poseControlModel`         | `control_v11p_sd15_openpose.safetensors` |
| `depthControlModel`        | `control_v11f1p_sd15_depth.safetensors`  |
| `segmentationControlModel` | `control_v11p_sd15_seg.safetensors`      |
| `poseStrength`             | `0.7`                                    |
| `depthStrength`            | `0.65`                                   |
| `segmentationStrength`     | `0.6`                                    |
| `batchSize`                | `1`                                      |
| `outputFormat`             | `png`                                    |
| `outputChannels`           | `4`                                      |

## Input Contract

| Input                      | Type    | Required | Default      |
| -------------------------- | ------- | -------- | ------------ |
| `pose_map`                 | image   | no       | n/a          |
| `depth_map`                | image   | no       | n/a          |
| `segmentation_map`         | image   | no       | n/a          |
| `use_pose_control`         | boolean | yes      | `true`       |
| `use_depth_control`        | boolean | yes      | `true`       |
| `use_segmentation_control` | boolean | yes      | `false`      |
| `pose_strength`            | float   | yes      | `0.7`        |
| `depth_strength`           | float   | yes      | `0.65`       |
| `segmentation_strength`    | float   | yes      | `0.6`        |
| `character_archetype`      | enum    | yes      | `hero`       |
| `style_preset`             | enum    | yes      | `stylized`   |
| `seed`                     | integer | yes      | `2026072704` |
| `width`                    | integer | yes      | `1024`       |
| `height`                   | integer | yes      | `1536`       |
| `steps`                    | integer | yes      | `35`         |
| `cfg_scale`                | float   | yes      | `6.4`        |

## Graph Nodes

| Node ID | Class                    | Role                       | Output Count |
| ------- | ------------------------ | -------------------------- | ------------ |
| `n1`    | `LoadImage`              | pose-control-input         | 2            |
| `n2`    | `LoadImage`              | depth-control-input        | 2            |
| `n3`    | `LoadImage`              | segmentation-control-input | 2            |
| `n4`    | `CheckpointLoaderSimple` | model-load                 | 3            |
| `n5`    | `ControlNetLoader`       | pose-control-load          | 1            |
| `n6`    | `ControlNetLoader`       | depth-control-load         | 1            |
| `n7`    | `ControlNetLoader`       | segmentation-control-load  | 1            |
| `n8`    | `CLIPTextEncode`         | positive-conditioning      | 1            |
| `n9`    | `CLIPTextEncode`         | negative-conditioning      | 1            |
| `n10`   | `ControlNetApply`        | pose-control-apply         | 1            |
| `n11`   | `ControlNetApply`        | depth-control-apply        | 1            |
| `n12`   | `ControlNetApply`        | segmentation-control-apply | 1            |
| `n13`   | `EmptyLatentImage`       | latent-init                | 1            |
| `n14`   | `KSampler`               | latent-sample              | 1            |
| `n15`   | `VAEDecode`              | image-decode               | 1            |
| `n16`   | `SaveImage`              | artifact-output            | 0            |

## Graph Edges

| Edge ID | From    | To                 |
| ------- | ------- | ------------------ |
| `e1`    | `n4:0`  | `n14.model`        |
| `e2`    | `n4:1`  | `n8.clip`          |
| `e3`    | `n4:1`  | `n9.clip`          |
| `e4`    | `n4:2`  | `n15.vae`          |
| `e5`    | `n8:0`  | `n10.conditioning` |
| `e6`    | `n5:0`  | `n10.control_net`  |
| `e7`    | `n1:0`  | `n10.image`        |
| `e8`    | `n10:0` | `n11.conditioning` |
| `e9`    | `n6:0`  | `n11.control_net`  |
| `e10`   | `n2:0`  | `n11.image`        |
| `e11`   | `n11:0` | `n12.conditioning` |
| `e12`   | `n7:0`  | `n12.control_net`  |
| `e13`   | `n3:0`  | `n12.image`        |
| `e14`   | `n12:0` | `n14.positive`     |
| `e15`   | `n9:0`  | `n14.negative`     |
| `e16`   | `n13:0` | `n14.latent_image` |
| `e17`   | `n14:0` | `n15.samples`      |
| `e18`   | `n15:0` | `n16.images`       |

## Topological Node Order

`n1` -> `n13` -> `n2` -> `n3` -> `n4` -> `n5` -> `n6` -> `n7` -> `n8` -> `n10`
-> `n11` -> `n12` -> `n9` -> `n14` -> `n15` -> `n16`

## Research Alignment

| Workflow                    | Path                                                                  | Control Nodes |
| --------------------------- | --------------------------------------------------------------------- | ------------- |
| AnimateDiff with ControlNet | `libs/isis/workflows/src/workflows/video/animatediff-controlnet.json` | 3             |
| Face Swap Avatar Generator  | `libs/isis/workflows/src/workflows/avatar/face-swap.json`             | 0             |

## Acceptance Criteria

- Graph is acyclic and topologically sortable.
- Pose/depth/segmentation control paths are explicitly represented through
  ControlNet loaders and apply nodes.
- At least one control mode can be enabled per run without breaking
  deterministic sampler/model defaults.
- Output contract aligns to MXA-0001 required output classes.

## Traceability

- Checklist item: `MXA-0004`
- Ticket key: `ISIS2-00202`
- Source references:
  - `docs/domains/isis/remediation-v2/ISIS_REMEDIATION_V2_TICKETS.json`
  - `docs/domains/isis/remediation-v2/ISIS_V2_CHARACTER_CONCEPT_ART_REQUIREMENTS_BRIEF.json`
  - `docs/domains/isis/remediation-v2/ISIS_V2_CHARACTER_CONCEPT_ART_BASELINE_WORKFLOW_GRAPH.json`
  - `docs/domains/isis/remediation-v2/ISIS_V2_CHARACTER_CONCEPT_ART_IMAGE_CONDITIONED_WORKFLOW_GRAPH.json`
  - `libs/isis/workflows/src/workflows/video/animatediff-controlnet.json`
  - `libs/isis/workflows/src/workflows/avatar/face-swap.json`
  - `docs/domains/isis/remediation-v2/ISIS_V2_CONTROL_TAXONOMY.json`
  - `ISIS_TODOS_2.md`
