contra-labs/firefly-creative-campaign-trajectories

Adobe Firefly Creative-Campaign Computer-Use Trajectories (Preview)

A preview release of long-horizon computer-use agent trajectories from creative-campaign work in Adobe Firefly (generative image and video), with reference-gathering in the browser. Each operator, a Designer sourced from Contra's top-earning talent carries a brand brief through to a set of campaign deliverables, evaluating each result as they go; every step pairs a screenshot with a structured action, an executable code string, and reasoning augmented from the operator's spoken narration, and every trajectory carries an expert quality evaluation (alignment, efficiency, difficulty, and a written rationale).

A sample of the human data Contra Labs builds on demand. We are an independent human-data and creative-evaluation lab: expert evaluation, rankings, and benchmarks for AI outputs, plus custom datasets like this one. See Working with Contra Labs.

Motivation

Most open computer-use and GUI-agent datasets focus on short, single-goal web or OS tasks. This preview is a small sample of a different kind of data: long-horizon generative-media workflows, where the operator iterates a brand brief across many prompt-and-generate cycles in Adobe Firefly, switches to the browser to gather references, and judges each deliverable against the brief. Alongside the trajectory, each episode carries an expert evaluation of how well the workflow met its goal.

What a session contains

A session follows the operator building a campaign in Adobe Firefly: pasting a brand brief, generating and refining a hero still, adapting it to additional aspect ratios (e.g. a 9:16 social variant), generating a short vertical video from the still, exporting assets, and recording an evaluation of each result. The operator works mostly in Firefly (generate, virtual try-on, edit, expand, export) with occasional reference-gathering in the browser. The five sampled campaigns are a fashion hero still + video (slow-fashion), an espresso-martini product set, a GPS-pendant brand launch, a slow-fashion editorial hero, and a "No Seed Oils" beanie hero. Some sessions span multiple segment-trajectories; group by session_uuid to recover a session.

Format and schema

The dataset is a single Parquet file with one row per step: 137 steps across 6 trajectories from 5 sessions (6 to 35 steps per trajectory). Screenshots are embedded in the Parquet as a Hugging Face Image feature. Trajectory-level fields are repeated on every step so a row is self-contained; group by trajectory_uuid and order by step_index to reconstruct a trajectory, and by session_uuid to group segment-trajectories from one recording session. Each trajectory ends with a terminate step marking the stop decision.

from datasets import load_dataset

ds = load_dataset("contra-labs/firefly-creative-campaign-trajectories", split="train")
row = ds[0]
print(row["instruction"])
print(row["action"], "->", row["code"])
row["image"].show()

Per-step fields:

Column	Description
trajectory_uuid	stable UUID for the trajectory (group on this to recover one segment-trajectory)
session_uuid	stable UUID for the recording session (group on this to recover all segments of a session)
step_index	position of the step within its trajectory (contiguous, 0-based)
image	the step's screenshot (Hugging Face Image feature)
thought	the step's reasoning
action, action_type	human-readable action and its taxonomy label (click, type, write, hotkey, navigate, terminate; type/write are both keyboard entry)
code	executable pyautogui-style action string (e.g. pyautogui.click(...), computer.terminate(...))
coord_x, coord_y	normalized pointer coordinates when applicable, otherwise null
observation, reflection	on-screen state before the action, and an assessment of the resulting state
viewport_width, viewport_height	screen dimensions for the step
timestamp_ms	step time offset

Trajectory-level fields (denormalized, repeated on every step):

Column	Description
instruction, natural_language_task, actual_task	the task stated at increasing levels of specificity
reason	the evaluator's written rationale for the scores
task_completed	whether the campaign deliverables and evaluation were completed
alignment_score, efficiency_score, task_difficulty	expert evaluation scores (alignment 6-8, efficiency 5, difficulty 6-8 in this sample)
step_count	number of steps in the trajectory

Reasoning provenance: thought, observation, and reflection are model-augmented from the operator's narration transcripts (the operator narrated each step aloud as they worked); action/code/coordinates and the screenshots are the operator's real execution; alignment_score / efficiency_score / task_difficulty / reason are the expert evaluation.

Relation to computer-use research

Each step follows the per-step pattern shared by computer-use corpora such as AgentNet (and related work like AgentTrek): a screenshot, a first-person observation and thought, a structured action (an action-type label plus a pyautogui-style action code), and a post-step reflection. Using a familiar shape keeps the data interoperable with existing computer-use tooling rather than introducing a bespoke schema. The additions specific to this set are generative-media tasks (Firefly image + video) and a per-trajectory expert evaluation (alignment_score, efficiency_score, task_difficulty, reason).

Curation

Each trajectory was cleaned to a single coherent path: low-value steps (misclicks, no-ops, abandoned exploratory branches) were dropped and indices renumbered contiguously, and every trajectory ends with a clean terminate step. Each step was reviewed against its screenshot so the action, observation, and reflection match what is on screen, with click coordinates checked against the controls they target and terminal and evaluation claims kept to what the captured frames show. Typed prompts were verified against the session's DOM recording (rrweb) so the recorded text matches what the operator entered, preserving original spelling. Step counts and per-step fields are kept consistent with the screenshots.

Intended use

• Exploring what long-horizon, multi-step generative-media trajectories look like as training and evaluation data.
• Studying creative computer-use workflows (prompt-iterate-evaluate loops) and how expert evaluation scores relate to the trajectory.
• Previewing the trajectory + evaluation schema for creative-professional work.

Limitations and scope

This is a preview rather than a training-scale corpus: 6 trajectories / 137 steps from 5 sessions in a single vertical (generative creative campaigns in Adobe Firefly). efficiency_score is constant in this sample, and the reasoning fields are model-augmented from the operator's narration rather than transcribed verbatim. Intended for evaluation and qualitative study rather than large-scale training.

Working with Contra Labs

Contra Labs is an independent human-data and creative-evaluation lab, backed by a network of verified creative and domain experts. This dataset is one example of our work.

We partner with AI teams on:

• Evaluation, rankings, and benchmarks. Expert human judgment on model outputs across text, image, video, audio, UI, and multi-modal work, scored for quality, style, and brand fit.
• Custom dataset creation. Computer-use trajectories, pairwise preference data, and evaluation scores, custom built to your domain, schema, and difficulty.

To commission an evaluation or dataset for your domain, reach out through contralabs.com or email partnerships@contralabs.com.

Provenance and consent

Recorded with consent from the participant. Screenshots were reviewed and the operator's own personally identifying information - account avatars, account name, and personal names in private app chrome - was redacted with an opaque marker. Stock, AI-generated, and third-party imagery was left intact, as it does not identify the operator: the AI-generated campaign deliverables on the canvas, sample thumbnails, and the brand, product, and tool names shown are all part of the work itself.

References

• AgentNet computer-use dataset: huggingface.co/datasets/xlangai/AgentNet. OpenCUA: Open Foundations for Computer-Use Agents, arXiv:2508.09123.
• AgentTrek: Agent Trajectory Synthesis via Guiding Replay with Web Tutorials. ICLR 2025. arXiv:2412.09605.

Citation

@misc{contra_labs_firefly_creative_campaign_trajectories_2026,
  title        = {Adobe Firefly Creative-Campaign Computer-Use Trajectories (Preview)},
  author       = {Contra Labs},
  year         = {2026},
  howpublished = {Hugging Face Datasets},
  note         = {Preview release}
}

License

Released under CC-BY-4.0. Free to use with attribution to Contra Labs.