diff --git a/README.md b/README.md
index f04630d..9dbd1d3 100644
--- a/README.md
+++ b/README.md
@@ -152,6 +152,7 @@ llm_backend = "auto"        # auto | api | claude-cli | gemini | openai
 max_turns = 60
 budget_usd = 5.0
 base_image = "readme2demo/base:latest"
+formats = ["demo", "gif"]
 skip_video = false
 ```
 
diff --git a/docs/configuration.md b/docs/configuration.md
index fb20e50..d9cdac6 100644
--- a/docs/configuration.md
+++ b/docs/configuration.md
@@ -18,6 +18,7 @@ llm_backend = "auto"        # auto | api | claude-cli | gemini | openai
 max_turns = 60
 budget_usd = 5.0
 base_image = "readme2demo/base:latest"
+formats = ["demo", "gif"]
 skip_video = false
 ```
 
diff --git a/docs/usage.md b/docs/usage.md
index 7a6c956..8ecf6f1 100644
--- a/docs/usage.md
+++ b/docs/usage.md
@@ -35,6 +35,7 @@ provide both. At least one is required.
 | `--anthropic [model]` | Run the sandboxed agent on OpenHands with a Claude model via `ANTHROPIC_API_KEY`. |
 | `--budget-usd` | Abort the run if the agent's cost exceeds this. |
 | `--dry-run` | Stop after ingest/planning with the feasibility verdict and blockers — no agent time spent. |
+| `--formats` | Comma-separated outputs to request (`demo,gif` today; others reserved). |
 | `--skip-video` / `--with-video` | Skip or force VHS rendering. |
 | `--allow-docker-socket` | Mount the host Docker socket into the sandbox. Security tradeoff — trusted repos only. |
 
diff --git a/src/readme2demo/cli.py b/src/readme2demo/cli.py
index 4800c46..fc1c17d 100644
--- a/src/readme2demo/cli.py
+++ b/src/readme2demo/cli.py
@@ -25,6 +25,7 @@
 from rich.markup import escape
 
 from readme2demo.config import Config
+from readme2demo.formats import FormatError, parse_formats
 from readme2demo.manifest import STAGES, Manifest
 from readme2demo.orchestrator import (
     Orchestrator,
@@ -116,6 +117,7 @@ def _build_config(
     budget_usd: Optional[float],
     max_turns: Optional[int],
     skip_video: Optional[bool],
+    formats: Optional[list[str]],
     base_image: Optional[str],
     llm_backend: Optional[str] = None,
 ) -> Config:
@@ -128,6 +130,7 @@ def _build_config(
         budget_usd=budget_usd,
         max_turns=max_turns,
         skip_video=skip_video,
+        formats=formats,
         base_image=base_image,
         llm_backend=llm_backend,
     )
@@ -310,6 +313,11 @@ def run(
     budget_usd: Optional[float] = typer.Option(None, help="Abort if agent cost exceeds this"),
     max_turns: Optional[int] = typer.Option(None, help="Agent max turns"),
     skip_video: Optional[bool] = typer.Option(None, "--skip-video/--with-video"),
+    formats: Optional[str] = typer.Option(
+        None,
+        "--formats",
+        help="Comma-separated output formats (demo,gif today; podcast/promo/social reserved).",
+    ),
     base_image: Optional[str] = typer.Option(None, help="Sandbox base image"),
     step_by_step: Optional[Path] = typer.Option(
         None, "-s", "--step-by-step",
@@ -379,9 +387,17 @@ def run(
             provider, engine, model, llm_backend, preset_model
         )
         _announce_preset(provider, model)
+
+    parsed_formats = None
+    if formats is not None:
+        try:
+            parsed_formats = parse_formats(formats)
+        except FormatError as e:
+            raise typer.BadParameter(str(e), param_hint="--formats") from e
+
     cfg = _build_config(
         config_file, engine, model, output_dir, timeout,
-        budget_usd, max_turns, skip_video, base_image, llm_backend,
+        budget_usd, max_turns, skip_video, parsed_formats, base_image, llm_backend,
     )
     if dry_run:
         cfg = cfg.model_copy(update={"dry_run": True})
@@ -401,6 +417,9 @@ def run(
         )
     cfg = _apply_engine_image(cfg)
     _preflight(cfg)
+    console.print(
+        f"[dim]Selected formats: {escape(', '.join(cfg.formats))}[/]"
+    )
     orch = Orchestrator.new_run(repo_url, cfg)
     _drive(orch)
 
diff --git a/src/readme2demo/config.py b/src/readme2demo/config.py
index 61bf7fc..1c3014c 100644
--- a/src/readme2demo/config.py
+++ b/src/readme2demo/config.py
@@ -7,7 +7,7 @@
 from pathlib import Path
 from typing import Any, Optional
 
-from pydantic import BaseModel, ConfigDict, Field, model_validator
+from pydantic import BaseModel, ConfigDict, Field, field_validator, model_validator
 
 if sys.version_info >= (3, 11):
     import tomllib
@@ -52,6 +52,9 @@ class Config(BaseModel):
     verify_retries: int = 1  # plain script retries before distiller feedback loop
     distill_retries: int = 1  # distiller feedback loops on verify failure
     skip_video: bool = False
+    # Selected output formats (registry in formats.py). Surface only in
+    # this slice — the pipeline still keys render off skip_video alone.
+    formats: list[str] = Field(default_factory=lambda: ["demo", "gif"])
 
     # Optional user-supplied step-by-step guide (-s/--step-by-step): injected
     # into the cloned repo so the planner and agent treat it as authoritative,
@@ -72,6 +75,14 @@ def _warn_deprecated_vhs_image(cls, data: Any) -> Any:
             )
         return data
 
+
+    @field_validator("formats", mode="after")
+    @classmethod
+    def _validate_formats(cls, value: list[str]) -> list[str]:
+        from readme2demo.formats import _validate_format_names
+
+        return _validate_format_names(list(value))
+
     @classmethod
     def load(cls, toml_path: Optional[Path] = None, **overrides: Any) -> "Config":
         """Build config from optional TOML file plus explicit overrides.
diff --git a/src/readme2demo/formats.py b/src/readme2demo/formats.py
new file mode 100644
index 0000000..8290d26
--- /dev/null
+++ b/src/readme2demo/formats.py
@@ -0,0 +1,107 @@
+"""Output-format registry and pure CLI/TOML parsing for multi-format selection.
+
+Surface-only slice of #117: declares which formats exist, which are implemented,
+and validates user selection. Does **not** change what the pipeline renders.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Optional
+
+
+class FormatError(ValueError):
+    """Raised when a formats selection is unknown or not yet implemented."""
+
+
+@dataclass(frozen=True)
+class FormatSpec:
+    """One selectable output format (row in the formats registry)."""
+
+    name: str
+    implemented: bool
+    description: str
+    # Tracking issue for unimplemented formats; shown in error messages.
+    tracking_issue: Optional[int] = None
+
+
+FORMATS: dict[str, FormatSpec] = {
+    "demo": FormatSpec(
+        name="demo",
+        implemented=True,
+        description="Primary demo video (demo.mp4)",
+    ),
+    "gif": FormatSpec(
+        name="gif",
+        implemented=True,
+        description="GIF preview of the demo (demo.gif)",
+    ),
+    "podcast": FormatSpec(
+        name="podcast",
+        implemented=False,
+        description="Podcast / audio narration of the demo",
+        tracking_issue=111,
+    ),
+    "promo": FormatSpec(
+        name="promo",
+        implemented=False,
+        description="Short promo cut of the demo video",
+        tracking_issue=114,
+    ),
+    "social": FormatSpec(
+        name="social",
+        implemented=False,
+        description="Social-media cut of the demo video",
+        tracking_issue=116,
+    ),
+}
+
+
+def parse_formats(value: str) -> list[str]:
+    """Parse a comma-separated formats string into a validated name list.
+
+    Splits on commas, strips whitespace, lowercases, drops empties, and
+    de-duplicates while preserving order. Raises :class:`FormatError` for
+    unknown names or declared-but-unimplemented formats.
+    """
+    if value is None:
+        raise FormatError("formats value is required")
+    raw_parts = [p.strip().lower() for p in value.split(",")]
+    names: list[str] = []
+    seen: set[str] = set()
+    for part in raw_parts:
+        if not part:
+            continue
+        if part in seen:
+            continue
+        seen.add(part)
+        names.append(part)
+    if not names:
+        raise FormatError(
+            "no formats specified; known formats: "
+            + ", ".join(sorted(FORMATS))
+        )
+    return _validate_format_names(names)
+
+
+def _validate_format_names(names: list[str]) -> list[str]:
+    """Validate a list of format names against the registry."""
+    implemented = sorted(n for n, s in FORMATS.items() if s.implemented)
+    known = sorted(FORMATS)
+    for name in names:
+        spec = FORMATS.get(name)
+        if spec is None:
+            raise FormatError(
+                f"unknown format {name!r}. Known formats: {', '.join(known)}"
+            )
+        if not spec.implemented:
+            issue = (
+                f" — tracked in #{spec.tracking_issue}"
+                if spec.tracking_issue
+                else ""
+            )
+            raise FormatError(
+                f"format {name!r} is not implemented yet{issue}. "
+                f"Implemented formats: {', '.join(implemented)}"
+            )
+    return names
diff --git a/tests/test_cli.py b/tests/test_cli.py
index 27e7a62..a7c4847 100644
--- a/tests/test_cli.py
+++ b/tests/test_cli.py
@@ -652,3 +652,71 @@ def test_report_json_and_markdown_are_mutually_exclusive(tmp_path):
     # A usage error, not silent precedence: neither format was emitted.
     assert "Verified" not in result.output
     assert '"stages"' not in result.output
+
+
+# -- formats registry surface (#192) ------------------------------------------
+
+
+def test_formats_option_accepted_and_echoed(tmp_path, monkeypatch):
+    """Regression: --formats demo,gif is accepted and echoed at startup."""
+    from readme2demo.orchestrator import Orchestrator
+    from readme2demo.manifest import Manifest
+
+    monkeypatch.setattr("readme2demo.cli._preflight", lambda cfg: None)
+
+    def fake_new_run(repo_url, cfg):
+        run_dir = tmp_path / "run"
+        m = Manifest.create(run_dir, repo_url or _URL, "claude-code", "img")
+        m.verified = True
+        m.save()
+        orch = Orchestrator.__new__(Orchestrator)
+        orch.run_dir = run_dir
+        orch.cfg = cfg
+        orch.manifest = m
+        return orch
+
+    def fake_run(self):
+        return self.manifest
+
+    monkeypatch.setattr(Orchestrator, "new_run", staticmethod(fake_new_run))
+    monkeypatch.setattr(Orchestrator, "run", fake_run)
+
+    result = runner.invoke(
+        app, ["run", _URL, "--formats", "DEMO,demo,gif", "--skip-video"]
+    )
+    assert result.exit_code == 0, result.output
+    assert "Selected formats:" in result.output
+    assert "demo" in result.output
+    assert "gif" in result.output
+
+
+def test_formats_unknown_is_bad_option_not_toml_error():
+    """Regression: bad --formats must not be reported as a toml config error."""
+    result = runner.invoke(app, ["run", _URL, "--formats", "banana"])
+    assert result.exit_code != 0
+    out = result.output
+    assert "unknown format" in out.lower() or "banana" in out
+    assert "readme2demo.toml" not in out
+
+
+def test_formats_unimplemented_mentions_tracking_issue():
+    """Regression: unimplemented formats say not implemented yet + issue number."""
+    result = runner.invoke(app, ["run", _URL, "--formats", "podcast"])
+    assert result.exit_code != 0
+    out = result.output.lower()
+    assert "not implemented" in out
+    assert "#111" in result.output or "111" in result.output
+    assert "unknown format" not in out
+
+
+def test_parse_formats_unit():
+    """Regression: parse_formats normalizes case, spaces, and de-duplicates."""
+    from readme2demo.formats import parse_formats, FormatError
+    import pytest
+
+    assert parse_formats("demo, gif") == ["demo", "gif"]
+    assert parse_formats("DEMO,demo,gif") == ["demo", "gif"]
+    with pytest.raises(FormatError, match="unknown"):
+        parse_formats("banana")
+    with pytest.raises(FormatError, match="not implemented"):
+        parse_formats("promo")
diff --git a/tests/test_config.py b/tests/test_config.py
index 8917390..6aec2cf 100644
--- a/tests/test_config.py
+++ b/tests/test_config.py
@@ -38,6 +38,7 @@ def test_defaults_without_any_toml(
         assert cfg.verify_retries == 1
         assert cfg.distill_retries == 1
         assert cfg.skip_video is False
+        assert cfg.formats == ["demo", "gif"]
         assert cfg.step_by_step is None
         assert cfg.runs_dir == Path("runs")
 
@@ -149,3 +150,33 @@ def test_unknown_override_kwarg_raises(
         monkeypatch.chdir(tmp_path)
         with pytest.raises(ValidationError, match="totally_unknown"):
             Config.load(totally_unknown="x")
+
+
+# --- formats registry (surface-only #192) -------------------------------------
+
+
+class TestFormats:
+    def test_formats_default(self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
+        monkeypatch.chdir(tmp_path)
+        assert Config.load().formats == ["demo", "gif"]
+
+    def test_formats_toml_accepted(self, tmp_path: Path) -> None:
+        toml = _write_toml(tmp_path / "r2d.toml", 'formats = ["demo"]\n')
+        assert Config.load(toml).formats == ["demo"]
+
+    def test_formats_toml_unknown_rejected(self, tmp_path: Path) -> None:
+        """Regression: bad toml formats must fail Config.load, not silent ignore."""
+        toml = _write_toml(tmp_path / "r2d.toml", 'formats = ["banana"]\n')
+        with pytest.raises(ValidationError, match="unknown format"):
+            Config.load(toml)
+
+    def test_formats_toml_unimplemented_rejected(self, tmp_path: Path) -> None:
+        """Regression: declared-but-unimplemented formats are not silently accepted."""
+        toml = _write_toml(tmp_path / "r2d.toml", 'formats = ["podcast"]\n')
+        with pytest.raises(ValidationError, match="not implemented yet"):
+            Config.load(toml)
+
+    def test_none_formats_flag_does_not_clobber_toml(self, tmp_path: Path) -> None:
+        """Regression: unpassed --formats (None) must keep toml formats."""
+        toml = _write_toml(tmp_path / "r2d.toml", 'formats = ["demo"]\n')
+        assert Config.load(toml, formats=None).formats == ["demo"]