Talent Direction is an Art Direction feature in Dreem Studio that lets you control your Virtual Model’s expression and mood. It complements Aesthetics (which sets background, lighting, and style) by defining how your model looks and feels, so the same expression stays consistent across poses and full catalog runs.

When you select a mood before generating, Dreem uses it to guide the model's expression as part of the output. Your talent, pose, and outfit selections work exactly as before. Talent Direction layers on top to define the demeanor your model shows.

Each mood is built from four expression cues, shown with every option:

Because the expression is set before you generate rather than adjusted afterward, you skip a separate editing step and get a more consistent look across a batch.

Talent Direction is now the only way to set a model's expression. Once it is available in your workspace, mood can no longer be changed through prompts or Image Editing. To adjust how a model looks, switch to a different mood in the Talent Direction library rather than describing the change in a prompt.

Prefer a quick walkthrough? Watch this short tutorial to see how Talent Direction works in Dreem:

You can also follow the step-by-step guide below.

In Dreem Studio, open the Art Direction panel and select Talent Direction to access the Curated Moods library. You'll find a range of ready-made expressions, each labelled with its four cues so you can see what you're selecting at a glance. Use the filters to narrow options by Eyes, Gaze, Head, Mouth, or who created them.

Once you find a mood that fits, hover it and select Use this Mood. It moves to the Mood Selected panel, where you can confirm the details before you commit. Then select Apply & Continue to bring it into your session.

If none of the library options match the look you want, you can create a custom mood by selecting "Create Your Own." Custom moods are saved under Your Moods, ready to reuse in future sessions.

Step 1: Match a reference

Upload a clear frame of the face and expression you want to reproduce. Dreem reads the expression parameters from it and gives you a starting point to refine.

Step 2: Review and adjust

Dreem fills in the mood from your reference. Fine-tune the fourcues (Eyes, Gaze, Head, and Mouth) to match your intent. Add Rules for anything that must hold, such as "Head stays level" or "Always into lens," and use Additional nuance for the small touches the cues don't cover, such as "Eyes wide open, brows slightly raised." Then click “Generate” so Dreem can re-analyze your updated inputs and show a live preview on a model as you go, so you can confirm the expression before committing.

Step 3: Save and reuse

When the preview matches what you had in mind, select Save. Your custom mood is then stored under Your Moods and available to select in any future session.

Talent Direction does not have to be set in just one place. You can apply it broadly or narrow it down to a single pose, and Dreem works out which setting to use. There are four levels:

(1) General, set in the Art Direction panel, acts as your workspace default

(2) Project, set on a project, applies one mood across every session in that project

(3) Talent, set on an individual model in the Talent Library, means that model always carries the expression you chose

(4) Pose, set on an individual pose in the Pose Library, gives you the most precise control. To set a mood on a talent or a pose, open its settings from the Talent Direction icon in the library.

When more than one level has a mood set, Dreem applies the most specific mood available, following this order:

It stops at the first mood it finds.

Example: If a Pose mood is set, it overrides the moods set at the Talent, Project, and General levels.

Only one level applies to any generation, and the levels override each other rather than combining. Where a level has no mood of its own, it follows the next level up, ultimately your General setting, shown in the panel as "Inherited from Global." Once you set a mood at any level, it stays there until you change it.

Selecting a mood does not change the rest of the Virtual Model workflow. You still choose your talent, pose, and style your outfit using the Outfit Builder as usual, and the mood that applies follows the level order above.

You do not need to start a new session to work with a different expression. Switch to a different Talent Direction, and any new generation you run follows the mood you just selected.

Run Again works a little differently. By default, it repeats the previous run, so it keeps the Talent Direction that run used rather than picking up a mood you have since changed at the General level. Each selected pose shows the mood it will use, marked "From previous run" when it is carrying the earlier setting. To have Run Again use a different expression, set the mood at the Pose level for the poses you want to change. That pose-level setting takes precedence, so the next run follows it.

This also gives you variety across a set. If you want most shots with an open-mouth smile but one with a closed-mouth smile, set the different mood at the Pose level for that shot, and leave the rest to follow your General or Talent setting.

Setting a mood at the General level keeps the same expression consistent across everything you generate, so a full catalog feels coherent without re-selecting your look each time. When a particular talent or pose needs its own expression, set it at that level and let it override the default just for those shots.

If your brand has a signature expression, for example a warm closed-mouth smile on every shot, saving it as a custom mood gives you a repeatable starting point without having to rebuild it for every shoot. Use Rules to lock the elements that must not change, and Additional nuance for the smaller adjustments, so the core expression stays stable while still leaving room for detail.

Remember that expression is controlled through Talent Direction alone. To change how a model looks, switch to a different mood rather than reaching for a prompt or Image Editing.